B | Perception, Vision, and Natural Language Processing

Road surface reconstruction from aerial images is fundamental for autonomous driving, urban planning, and virtual simulation, where smoothness, compactness, and accuracy are critical quality factors. Existing reconstruction methods often produce artifacts and inconsistencies that limit usability, while downstream tasks have a tendency to represent roads as planes for simplicity but at the cost of accuracy. We introduce FlexRoad, the first framework to directly address road surface smoothing by fitting Non-Uniform Rational B-Splines (NURBS) surfaces to 3D road points obtained from photogrammetric reconstructions or geodata providers. Our method at its core utilizes the Elevation-Constrained Spatial Road Clustering (ECSRC) algorithm for robust anomaly correction, significantly reducing surface roughness and fitting errors. To facilitate quantitative comparison between road surface reconstruction methods, we present GeoRoad Dataset (GeRoD), a diverse collection of road surface and terrain profiles derived from openly accessible geodata. Experiments on GeRoD and the photogrammetry-based DeepScenario Open 3D Dataset (DSC3D) demonstrate that FlexRoad considerably surpasses commonly used road surface representations across various metrics while being insensitive to various input sources, terrains, and noise types. By performing ablation studies, we identify the key role of each component towards high-quality reconstruction performance, making FlexRoad a generic method for realistic road surface modeling.

Accurate 3D trajectory data is crucial for advancing autonomous driving. Yet, traditional datasets are usually captured by fixed sensors mounted on a car and are susceptible to occlusion. Additionally, such an approach can precisely reconstruct the dynamic environment in the close vicinity of the measurement vehicle only, while neglecting objects that are further away. In this paper, we introduce the DeepScenario Open 3D Dataset (DSC3D), a high-quality, occlusion-free dataset of 6 degrees of freedom bounding box trajectories acquired through a novel monocular camera drone tracking pipeline. Our dataset includes more than 175,000 trajectories of 14 types of traffic participants and significantly exceeds existing datasets in terms of diversity and scale, containing many unprecedented scenarios such as complex vehicle-pedestrian interaction on highly populated urban streets and comprehensive parking maneuvers from entry to exit. DSC3D dataset was captured in five various locations in Europe and the United States and include: a parking lot, a crowded inner-city, a steep urban intersection, a federal highway, and a suburban intersection. Our 3D trajectory dataset aims to enhance autonomous driving systems by providing detailed environmental 3D representations, which could lead to improved obstacle interactions and safety. We demonstrate its utility across multiple applications including motion prediction, motion planning, scenario mining, and generative reactive traffic agents. Our interactive online visualization platform and the complete dataset are publicly available at this https URL, facilitating research in motion prediction, behavior modeling, and safety validation.

Scaling by training on large datasets has been shown to enhance the quality and fidelity of image generation and manipulation with diffusion models; however, such large datasets are not always accessible in medical imaging due to cost and privacy issues, which contradicts one of the main applications of such models to produce synthetic samples where real data is scarce. Also, finetuning on pre-trained general models has been a challenge due to the distribution shift between the medical domain and the pre-trained models. Here, we propose Latent Drift (LD) for diffusion models that can be adopted for any fine-tuning method to mitigate the issues faced by the distribution shift or employed in inference time as a condition. Latent Drifting enables diffusion models to be conditioned for medical images fitted for the complex task of counterfactual image generation, which is crucial to investigate how parameters such as gender, age, and adding or removing diseases in a patient would alter the medical images. We evaluate our method on three public longitudinal benchmark datasets of brain MRI and chest X-rays for counterfactual image generation. Our results demonstrate significant performance gains in various scenarios when combined with different fine-tuning schemes. The source code of this work will be publicly released upon its acceptance.

Dimensionality reduction is a fundamental task that aims to simplify complex data by reducing its feature dimensionality while preserving essential patterns, with core applications in data analysis and visualisation. To preserve the underlying data structure, multi-dimensional scaling (MDS) methods focus on preserving pairwise dissimilarities, such as distances. They optimise the embedding to have pairwise distances as close as possible to the data dissimilarities. However, the current standard is limited to embedding data in Riemannian manifolds. Motivated by the lack of asymmetry in the Riemannian metric of the embedding space, this paper extends the MDS problem to a natural asymmetric generalisation of Riemannian manifolds called Finsler manifolds. Inspired by Euclidean spaces, we define a canonical Finsler space for embedding asymmetric data. Due to its simplicity with respect to geodesics, data representation in this space is both intuitive and simple to analyse. We demonstrate that our generalisation benefits from the same theoretical convergence guarantees. We reveal the effectiveness of our Finsler embedding across various types of non-symmetric data, highlighting its value in applications such as data visualisation, dimensionality reduction, directed graph embedding, and link prediction.

Open-vocabulary semantic segmentation models associate vision and text to label pixels from an undefined set of classes using textual queries, providing versatile performance on novel datasets. However, large shifts between training and test domains degrade their performance, requiring fine-tuning for effective real-world application. We introduce Semantic Library Adaptation (SemLa), a novel framework for training-free, test-time domain adaptation. SemLa leverages a library of LoRA-based adapters indexed with CLIP embeddings, dynamically merging the most relevant adapters based on proximity to the target domain in the embedding space. This approach constructs an ad-hoc model tailored to each specific input without additional training. Our method scales efficiently, enhances explainability by tracking adapter contributions, and inherently protects data privacy, making it ideal for sensitive applications. Comprehensive experiments on an 18-domain benchmark built over 10 standard datasets demonstrate SemLa’s superior adaptability and performance across diverse settings, establishing a new standard in domain adaptation for open-vocabulary semantic segmentation.

In this work we address various shape matching problems that can be cast as finding cyclic paths in a product graph. This involves for example 2D-3D shape matching, 3D shape matching, or the matching of a contour to a graph. In this context, matchings are typically obtained as the minimum cost cycle in the product graph. Instead, inspired by related works on model-based image segmentation, we consider minimum ratio cycles, which we combine with the recently introduced conjugate product graph in order to allow for higher-order matching costs. With that, on the one hand we avoid the bias of obtaining matchings that involve fewer/shorter edges, while on the other hand being able to impose powerful geometric regularisation, e.g. to avoid zig-zagging. In our experiments we demonstrate that this not only leads to improved matching accuracy in most cases, but also to significantly reduced runtimes (up to two orders of magnitude, depending on the setting). Our GPU implementation will be made publicly available upon acceptance.

Recent advancements in diffusion models have established new benchmarks in both generative tasks and downstream applications. In contrast, flow matching models have shown promising improvements in performance but have not been as extensively explored, particularly due to the difficulty of inheriting knowledge from a pretrained diffusion prior foundation model.In this work, we propose a novel method to bridge the gap between pretrained diffusion models and flow matching models by aligning their trajectories and matching their objectives. Our approach mathematically formalizes this alignment and enables the efficient transfer of knowledge from diffusion priors to flow matching models. We demonstrate that our method outperforms traditional diffusion and flow matching finetuning, achieving competitive results across a variety of tasks.

We tackle the problem of automatic calibration of radially distorted cameras in challenging conditions.Accurately determining distortion parameters typically requires either 1) solving the full Structure from Motion (SfM) problem involving camera poses, 3D points, and the distortion parameters, which is only possible if many images with sufficient overlap are provided, or 2) relying heavily on learning-based methods that are comparatively less accurate.In this work, we demonstrate that distortion calibration can be decoupled from 3D reconstruction, maintaining the accuracy of SfM-based methods while avoiding many of the associated complexities. This is achieved by working in Projective Space, where the geometry is unique up to a homography, which encapsulates all camera parameters except for distortion.Our proposed method, Projective Radial Distortion Averaging, averages multiple distortion estimates in a fully projective framework without creating 3d points and full bundle adjustment. By relying on pairwise projective relations, our methods support any feature-matching approaches without constructing point tracks across multiple images.

Finding correspondences between 3D shapes is a crucial problem in computer vision and graphics. While most research has focused on finding correspondences in settings where at least one of the shapes is complete, the realm of partial-to-partial shape matching remains under-explored. Yet it is of importance since, in many applications, shapes are only observed partially due to occlusion or scanning.Finding correspondences between partial shapes comes with an additional challenge: We not only want to identify correspondences between points on either shape but also have to determine which points of each shape actually have a partner.To tackle this challenging problem, we present EchoMatch, a novel framework for partial-to-partial shape matching that incorporates the concept of correspondence reflection to enable an overlap prediction within a functional map framework.With this approach, we show that we can outperform current SOTA methods in challenging partial-to-partial shape matching problems.

Cooperative perception can increase the view field and decrease the occlusion of an ego vehicle, hence improving the perception performance and safety of autonomous driving. Despite the success of previous works on cooperative object detection, they mostly operate on dense Bird’s Eye View (BEV) feature maps, which is computationally demanding and can hardly be extended to long-range detection problems. More efficient fully sparse frameworks are rarely explored. In this work, we design a fully sparse framework, textit{SparseAlign}, with three key features: an enhanced sparse 3D backbone, a query-based temporal context learning module, and a robust detection head specially tailored for sparse features. Extensive experimental results on both OPV2V and DairV2X datasets show that our framework, despite sparsity, outperforms the state of the art with less communication bandwidth requirements. In addition, experiments on the OPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also show a significant performance gain compared to the baseline works.

In recent years, advances in text-to-image (T2I) diffusion models have substantially elevated the quality of their generated images. However, achieving fine-grained control over attributes remains a challenge due to the limitations of natural language prompts (such as no continuous set of intermediate descriptions existing between person'' and old person’’). Even though many methods were introduced that augment the model or generation process to enable such control, methods that do not require a fixed reference image are limited to either enabling global fine-grained attribute expression control or coarse attribute expression control localized to specific subjects, not both simultaneously. We show that there exist directions in the commonly used token-level CLIP text embeddings that enable fine-grained subject-specific control of high-level attributes in text-to-image models. Based on this observation, we introduce one efficient optimization-free and one robust optimization-based method to identify these directions for specific attributes from contrastive text prompts. We demonstrate that these directions can be used to augment the prompt text input with fine-grained control over attributes of specific subjects in a compositional manner (control over multiple attributes of a single subject) without having to adapt the diffusion model.

Probabilistic human motion prediction aims to forecast multiple possible future movements from past observations. While current approaches report high diversity and realism, they often generate motions with undetected limb stretching and jitter. To address this, we introduce SkeletonDiffusion, a latent diffusion model that embeds an explicit inductive bias on the human body within its architecture and training. Our model is trained with a novel nonisotropic Gaussian diffusion formulation that aligns with the natural kinematic structure of the human skeleton. Results show that our approach outperforms conventional isotropic alternatives, consistently generating realistic predictions while avoiding artifacts such as limb distortion. Additionally, we identify a limitation in commonly used diversity metrics, which may inadvertently favor models that produce inconsistent limb lengths within the same sequence. SkeletonDiffusion sets a new benchmark on three real-world datasets, outperforming various baselines across multiple evaluation metrics.

Model merging combines multiple expert models - finetuned from a base foundation model on diverse tasks and domains - into a single, more capable model. However, most existing model merging approaches assume that all experts are available simultaneously. In reality, new tasks and domains emerge progressively over time, requiring strategies to integrate the knowledge of expert models as they become available: a process we call temporal model merging. The temporal dimension introduces unique challenges not addressed in prior work, raising new questions such as: when training for a new task, should the expert model start from the merged past experts or from the original base model? Should we merge all models at each time step? Which merging techniques are best suited for temporal merging? Should different strategies be used to initialize the training and deploy the model? To answer these questions, we propose a unified framework called TIME - Temporal Integration of Model Expertise - which defines temporal model merging across three axes: (1) Initialization Phase, (2) Deployment Phase, and (3) Merging Technique. Using TIME, we study temporal model merging across model sizes, compute budgets, and learning horizons on the FoMo-in-Flux benchmark. Our comprehensive suite of experiments across TIME allows us to uncover key insights for temporal model merging, offering a better understanding of current challenges and best practices for effective temporal model merging.

Unsupervised panoptic segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training on manually annotated data. In contrast to prior work on unsupervised panoptic scene understanding, we eliminate the need for object-centric training data, enabling the unsupervised understanding of complex scenes. To that end, we present the first unsupervised panoptic method that directly trains on scene-centric imagery. In particular, we propose an approach to obtain high-resolution panoptic pseudo labels on complex scene-centric data combining visual representations, depth, and motion cues. Utilizing both pseudo-label training and a panoptic self-training strategy yields a novel approach that accurately predicts panoptic segmentation of complex scenes without requiring any human annotations. Our approach significantly improves panoptic quality, e.g., surpassing the recent state of the art in unsupervised panoptic segmentation on Cityscapes by 9.4% points in PQ.

Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of contrastive loss makes VLMs focus predominantly on foreground objects, neglecting other crucial information in the image, which limits their effectiveness in downstream tasks. To address these challenges, we propose COSMOS: CrOSs-MOdality Self-distillation for vision-language pre-training that integrates a novel text-cropping strategy and cross-attention module into a self-supervised learning framework. We create global and local views of images and texts (i.e., multi-modal augmentations), which are essential for self-distillation in VLMs. We further introduce a cross-attention module, enabling COSMOS to learn comprehensive cross-modal representations optimized via a cross-modality self-distillation loss. COSMOS consistently outperforms previous strong baselines on various zero-shot downstream tasks, including retrieval, classification, and semantic segmentation. Additionally, it surpasses CLIP-based models trained on larger datasets in visual perception and contextual understanding tasks.

Large-scale multimodal representation learning successfully optimizes for zero-shot transfer at test time. Yet the standard pretraining paradigm (contrastive learning on large amounts of image-text data) does not explicitly encourage representations to support few-shot adaptation. In this work, we propose a simple, but carefully designed extension to multimodal pretraining which enables representations to accommodate additional context. Using this objective, we show that vision-language models can be trained to exhibit significantly increased few-shot adaptation: across 21 downstream tasks, we find up to four-fold improvements in test-time sample efficiency, and average few-shot adaptation gains of over 5%, while retaining zero-shot generalization performance across model scales and training durations. In particular, equipped with simple, training-free, metric-based adaptation mechanisms, our representations easily surpass more complex and expensive optimization-based schemes, vastly simplifying generalization to new domains.

Generating realistic intermediate shapes between non-rigidly deformed shapes is a challenging task in computer vision, especially with unstructured data (e.g., point clouds) where temporal consistency across frames is lacking, and topologies are changing. Most interpolation methods are designed for structured data (i.e., meshes) and do not apply to real-world point clouds. In contrast, our approach, 4Deform, leverages neural implicit representation (NIR) to enable free topology changing shape deformation. Unlike previous mesh-based methods that learn vertex-based deformation fields, our method learns a continuous velocity field in Euclidean space. Thus, it is suitable for less structured data such as point clouds. Additionally, our method does not require intermediate-shape supervision during training; instead, we incorporate physical and geometrical constraints to regularize the velocity field. We reconstruct intermediate surfaces using a modified level-set equation, directly linking our NIR with the velocity field. Experiments show that our method significantly outperforms previous NIR approaches across various scenarios (e.g., noisy, partial, topology-changing, non-isometric shapes) and, for the first time, enables new applications like 4D Kinect sequence upsampling and real-world high-resolution mesh deformation.

The platonic representation hypothesis suggests that vision and language embeddings become more homogeneous as model and dataset sizes increase. In particular, pairwise distances within each modality become more similar. This suggests that as foundation models mature, it may become possible to match vision and language embeddings in a fully unsupervised fashion, i.e., without parallel data. We present the first study towards this prospect, and investigate conformity of existing vision and language foundation models in the context of ‘blind’ matching. First, we formulate unsupervised matching as a quadratic assignment problem and introduce a novel heuristic that outperforms previous solvers. We also develop a technique to find optimal matching problems, for which a non-trivial match is very likely. Second, we conduct an extensive study deploying a range of vision and language models on four datasets. Our analysis reveals that for many problem instances, vision and language representations can be indeed matched without supervision. This finding opens possibility for exciting applications embedding semantic knowledge into other modalities. As a showcase, we demonstrate a proof-of-concept unsupervised classifier, which achieves non-trivial classification accuracy without any image-text annotation.

Internal features from large-scale pre-trained diffusion models have recently been established as powerful semantic descriptors for a wide range of downstream tasks. Works that use these features generally need to add noise to images before passing them through the model to obtain the semantic features, as the models do not offer the most useful features when given images with little to no noise. We show that this noise has a critical impact on the usefulness of these features that cannot be remedied by ensembling with different random noises. We address this issue by introducing a lightweight, unsupervised fine-tuning method that enables diffusion backbones to provide high-quality, noise-free semantic features. We show that these features readily outperform previous diffusion features by a wide margin in a wide variety of extraction setups and downstream tasks, offering better performance than even ensemble-based methods at a fraction of the cost.

Estimating camera motion and intrinsics from casual videos is a core challenge in computer vision. Traditional bundle-adjustment based methods, such as SfM and SLAM, struggle to perform reliably on arbitrary data. Although specialized SfM approaches have been developed for handling dynamic scenes, they either require intrinsics or computationally expensive test-time optimization and often fall short in performance. Recently, methods like Dust3r have reformulated the SfM problem in a more data-driven way. While such techniques show promising results, they are still 1) not robust towards dynamic objects and 2) require labeled data for supervised training.As an alternative, we propose AnyCam, a fast transformer model that directly estimates camera poses and intrinsics from a dynamic video sequence in feed-forward fashion. Our intuition is that such a network can learn strong priors over realistic camera motions. To scale up our training, we rely on an uncertainty-based loss formulation and pre-trained depth and flow networks instead of motion or trajectory supervision. This allows us to use diverse, unlabelled video datasets obtained mostly from YouTube. Additionally, we ensure that the predicted trajectory does not accumulate drift over time through a lightweight trajectory refinement step. We test AnyCam on established datasets, where it delivers accurate camera poses and intrinsics both qualitatively and quantitatively. Furthermore, even with trajectory refinement, AnyCam is significantly faster than existing works for SfM in dynamic settings. Finally, by combining camera information, uncertainty, and depth, our model can produce high-quality 4D pointclouds in a feed-forward fashion.

CLIP has shown impressive results in aligning images and texts at scale. However, its ability to capture detailed visual features remains limited because CLIP matches images and texts at a global level. To address this issue, we propose FLAIR, Fine-grained Language-informed Image Representations, an approach that utilizes long and detailed image descriptions to learn localized image embeddings. By sampling diverse sub-captions that describe fine-grained details about an image, we train our vision-language model to produce not only global embeddings but also text-specific image representations. Our model introduces text-conditioned attention pooling on top of local image tokens to produce fine-grained image representations that excel at retrieving detailed image content. We achieve state-of-the-art performance on both, existing multimodal retrieval benchmarks, as well as, our newly introduced fine-grained retrieval task which evaluates vision-language models’ ability to retrieve partial image content. Furthermore, our experiments demonstrate the effectiveness of FLAIR trained on 30M image-text pairs in capturing fine-grained visual information, including zero-shot semantic segmentation, outperforming models trained on billions of pairs.

Video understanding is a pivotal task in the digital era, yet the dynamic and multievent nature of videos makes them labor-intensive and computationally demanding to process. Thus, localizing a specific event given a semantic query has gained importance in both user-oriented applications like video search and academic research into video foundation models. A significant limitation in current research is that semantic queries are typically in natural language that depicts the semantics of the target event. This setting overlooks the potential for multimodal semantic queries composed of images and texts. To address this gap, we introduce a new benchmark, ICQ, for localizing events in videos with multimodal queries, along with a new evaluation dataset ICQ-Highlight. Our new benchmark aims to evaluate how well models can localize an event given a multimodal semantic query that consists of a reference image, which depicts the event, and a refinement text to adjust the images’ semantics. To systematically benchmark model performance, we include 4 styles of reference images and 5 types of refinement texts, allowing us to explore model performance across different domains. We propose 3 adaptation methods that tailor existing models to our new setting and evaluate 10 SOTA models, ranging from specialized to large-scale foundation models. We believe this benchmark is an initial step toward investigating multimodal queries in video event localization.

Semantic 3D city models are worldwide easy-accessible, providing accurate, object-oriented, and semantic-rich 3D priors. To date, their potential to mitigate the noise impact on radar object detection remains under-explored. In this paper, we first introduce a unique dataset, RadarCity, comprising 54K synchronized radar-image pairs and semantic 3D city models. Moreover, we propose a novel neural network, RADLER, leveraging the effectiveness of contrastive self-supervised learning (SSL) and semantic 3D city models to enhance radar object detection of pedestrians, cyclists, and cars. Specifically, we first obtain the robust radar features via a SSL network in the radar-image pretext task. We then use a simple yet effective feature fusion strategy to incorporate semantic-depth features from semantic 3D city models. Having prior 3D information as guidance, RADLER obtains more fine-grained details to enhance radar object detection. We extensively evaluate RADLER on the collected RadarCity dataset and demonstrate average improvements of 5.46% in mean avarage precision (mAP) and 3.51% in mean avarage recall (mAR) over previous radar object detection methods. We believe this work will foster further research on semantic-guided and map-supported radar object detection.

In this paper, we propose MA-DV2F: Multi-Agent Dynamic Velocity Vector Field. It is a framework for simultaneously controlling a group of vehicles in challenging environments. DV2F is generated for each vehicle independently and provides a map of reference orientation and speed that a vehicle must attain at any point on the navigation grid such that it safely reaches its target. The field is dynamically updated depending on the speed and proximity of the ego-vehicle to other agents. This dynamic adaptation of the velocity vector field allows prevention of imminent collisions. Experimental results show that MA-DV2F outperforms concurrent methods in terms of safety, computational efficiency and accuracy in reaching the target when scaling to a large number of vehicles.

The generation of high-fidelity view synthesis is essential for robotic navigation and interaction but remains challenging, particularly in indoor environments and real-time scenarios. Existing techniques often require significant computational resources for both training and rendering, and they frequently result in suboptimal 3D representations due to insufficient geometric structuring. To address these limitations, we introduce VoxNeRF, a novel approach that utilizes easy-to-obtain geometry priors to enhance both the quality and efficiency of neural indoor reconstruction and novel view synthesis. We propose an efficient voxel-guided sampling technique that allocates computational resources selectively to the most relevant segments of rays based on a voxel-encoded geometry prior, significantly reducing training and rendering time. Additionally, we incorporate a robust depth loss to improve reconstruction and rendering quality in sparse view settings. Our approach is validated with extensive experiments on ScanNet and ScanNet++ where VoxNeRF outperforms existing state-of-the-art methods and establishes a new benchmark for indoor immersive interpolation and extrapolation settings.

We tackle the problem of monocular 3D object detection across different sensors, environments, and camera setups. In this paper, we introduce a novel unsupervised domain adaptation approach, MonoCT, that generates highly accurate pseudo labels for self-supervision. Inspired by our observation that accurate depth estimation is critical to mitigating domain shifts, MonoCT introduces a novel Generalized Depth Enhancement (GDE) module with an ensemble concept to improve depth estimation accuracy. Moreover, we introduce a novel Pseudo Label Scoring (PLS) module by exploring inner-model consistency measurement and a Diversity Maximization (DM) strategy to further generate high-quality pseudo labels for self-training. Extensive experiments on six benchmarks show that MonoCT outperforms existing SOTA domain adaptation methods by large margins (~21% minimum for AP Mod.) and generalizes well to car, traffic camera and drone views.

Continual learning (CL) aims to learn a sequence of tasks without forgetting prior knowledge, but gradient updates for a new task often overwrite the weights learned earlier, causing catastrophic forgetting (CF). We propose Prototype-Augmented Hypernetworks (PAH), a framework where a single hypernetwork, conditioned on learnable task prototypes, dynamically generates task-specific classifier heads on demand. To mitigate forgetting, PAH combines cross-entropy with dual distillation losses, one to align logits and another to align prototypes, ensuring stable feature representations across tasks. Evaluations on Split-CIFAR100 and TinyImageNet demonstrate that PAH achieves state-of-the-art performance, reaching 74.5% and 63.7% accuracy with only 1.7% and 4.4% forgetting, respectively, surpassing prior methods without storing samples or heads.

Standard graph neural networks assign vastly different latent embeddings to graphs describing the same object at different resolution scales. This precludes consistency in applications and prevents generalization between scales as would fundamentally be needed e.g. in AI4Science. We uncover the underlying obstruction, investigate its origin and show how to overcome it by modifying the message passing paradigm.

Diffusion models break down the challenging task of generating data from high-dimensional distributions into a series of easier denoising steps. Inspired by this paradigm, we propose a novel approach that extends the diffusion framework into modality space, decomposing the complex task of RGB image generation into simpler, interpretable stages. Our method, termed ToddlerDiffusion, cascades modality-specific models, each responsible for generating an intermediate representation, such as contours, palettes, and detailed textures, ultimately culminating in a high-quality RGB image. Instead of relying on the naive LDM concatenation conditioning mechanism to connect the different stages together, we employ Schrödinger Bridge to determine the optimal transport between different modalities. Although employing a cascaded pipeline introduces more stages, which could lead to a more complex architecture, each stage is meticulously formulated for efficiency and accuracy, surpassing Stable-Diffusion (LDM) performance. Modality composition not only enhances overall performance but enables emerging proprieties such as consistent editing, interaction capabilities, high-level interpretability, and faster convergence and sampling rate. Extensive experiments on diverse datasets, including LSUN-Churches, ImageNet, CelebHQ, and LAION-Art, demonstrate the efficacy of our approach, consistently outperforming state-of-the-art methods. For instance, ToddlerDiffusion achieves notable efficiency, matching LDM performance on LSUN-Churches while operating 2× faster with a 3× smaller architecture.

The physics solvers employed for neural network training are primarily iterative, and hence, differentiating through them introduces a severe computational burden as iterations grow large. Inspired by works in bilevel optimization, we show that full accuracy of the network is achievable through physics significantly coarser than fully converged solvers. We propose Progressively Refined Differentiable Physics (PRDP), an approach that identifies the level of physics refinement sufficient for full training accuracy. By beginning with coarse physics, adaptively refining it during training, and stopping refinement at the level adequate for training, it enables significant compute savings without sacrificing network accuracy. Our focus is on differentiating iterative linear solvers for sparsely discretized differential operators, which are fundamental to scientific computing. PRDP is applicable to both unrolled and implicit differentiation. We validate its performance on a variety of learning scenarios involving differentiable physics solvers such as inverse problems, autoregressive neural emulators, and correction-based neural-hybrid solvers. In the challenging example of emulating the Navier-Stokes equations, we reduce training time by 62%.

Parameter Efficient FineTuning (PEFT) methods have recently gained extreme popularity thanks to the vast availability of large-scale models, allowing to quickly adapt pretrained models to downstream tasks with minimal computational costs. However, current additive finetuning methods such as LoRA show low robustness to prolonged training and hyperparameter choices, not allowing for optimal out-of-the-box usage. On the other hand, multiplicative and bounded approaches such as ETHER, even if providing higher robustness, only allow for extremely low-rank adaptations and are limited to a fixed-strength transformation, hindering the expressive power of the adaptation. In this work, we propose the DeLoRA finetuning method that first normalizes and then scales the learnable low-rank matrices, thus effectively bounding the transformation strength, which leads to increased hyperparameter robustness at no cost in performance. We show that this proposed approach effectively and consistently improves over popular PEFT methods by evaluating our method on two finetuning tasks, subject-driven image generation and LLM instruction tuning.

Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved predictive performance, Mixup is also a good technique for improving calibration. However, mixing data carelessly can lead to manifold mismatch, i.e., synthetic data lying outside original class manifolds, which can deteriorate calibration. In this work, we show that the likelihood of assigning a wrong label with mixup increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves predictive performance and calibration of models, while being much more efficient.

Pre-trained large language models (LLMs) have been reliably integrated with visual input for multimodal tasks. The widespread adoption of instruction-tuned image-to-text vision-language assistants (VLAs) like LLaVA and InternVL necessitates evaluating gender biases. We study gender bias in 22 popular open-source VLAs with respect to personality traits, skills, and occupations. Our results show that VLAs replicate human biases likely present in the data, such as real-world occupational imbalances. Similarly, they tend to attribute more skills and positive personality traits to women than to men, and we see a consistent tendency to associate negative personality traits with men. To eliminate the gender bias in these models, we find that finetuning-based debiasing methods achieve the best tradeoff between debiasing and retaining performance on downstream tasks. We argue for pre-deploying gender bias assessment in VLAs and motivate further development of debiasing strategies to ensure equitable societal outcomes.

In this work, we introduce the first unsupervised method that simultaneously predicts time-varying neural implicit surfaces and deformations between pairs of point clouds. We propose to model the point movement using an explicit velocity field and directly deform a time-varying implicit field using the modified level-set equation. This equation utilizes an iso-surface evolution with Eikonal constraints in a compact formulation, ensuring the integrity of the signed distance field. By applying a smooth, volume-preserving constraint to the velocity field, our method successfully recovers physically plausible intermediate shapes. Our method is able to handle both rigid and non-rigid deformations without any intermediate shape supervision. Our experimental results demonstrate that our method significantly outperforms existing works, delivering superior results in both quality and efficiency.

State-of-the-art Denoising Diffusion Probabilistic Models (DDPMs) rely on an expensive sampling process with a large Number of Function Evaluations (NFEs) to provide high-fidelity predictions. This computational bottleneck renders diffusion models less appealing as surrogates for the spatio-temporal prediction of physics-based problems with long rollout horizons. We propose Truncated Sampling Models, enabling single-step and few-step sampling with elevated fidelity by simple truncation of the diffusion process, reducing the gap between DDPMs and deterministic single-step approaches. We also introduce a novel approach, Iterative Refinement, to sample pre-trained DDPMs by reformulating the generative process as a refinement process with few sampling steps. Both proposed methods enable significant improvements in accuracy compared to DDPMs, DDIMs, and EDMs with NFEs ≤ 10 on a diverse set of experiments, including incompressible and compressible turbulent flow and airfoil flow uncertainty simulations. Our proposed methods provide stable predictions for long rollout horizons in time-dependent problems and are able to learn all modes of the data distribution in steady-state problems with high uncertainty.

Learning disentangled representations from unlabelled data is a fundamental challenge in machine learning. Solving it may unlock other problems, such as generalization, interpretability, or fairness. Although remarkably challenging to solve in theory, disentanglement is often achieved in practice through prior matching. Furthermore, recent works have shown that prior matching approaches can be enhanced by leveraging geometrical considerations, e.g., by learning representations that preserve geometric features of the data, such as distances or angles between points. However, matching the prior while preserving geometric features is challenging, as a mapping that fully preserves these features while aligning the data distribution with the prior does not exist in general. To address these challenges, we introduce a novel approach to disentangled representation learning based on quadratic optimal transport. We formulate the problem using Gromov-Monge maps that transport one distribution onto another with minimal distortion of predefined geometric features, preserving them as much as can be achieved. To compute such maps, we propose the Gromov-Monge-Gap (GMG), a regularizer quantifying whether a map moves a reference distribution with minimal geometry distortion. We demonstrate the effectiveness of our approach for disentanglement across four standard benchmarks, outperforming other methods leveraging geometric considerations.

Masked transformer models for class-conditional image generation have become a compelling alternative to diffusion models. Typically comprising two stages - an initial VQGAN model for transitioning between latent space and image space, and a subsequent Transformer model for image generation within latent space - these frameworks offer promising avenues for image synthesis. In this study, we present two primary contributions: Firstly, an empirical and systematic examination of VQGANs, leading to a modernized VQGAN. Secondly, a novel embedding-free generation network operating directly on bit tokens - a binary quantized representation of tokens with rich semantics. The first contribution furnishes a transparent, reproducible, and high-performing VQGAN model, enhancing accessibility and matching the performance of current state-of-the-art methods while revealing previously undisclosed details. The second contribution demonstrates that embedding-free image generation using bit tokens achieves a new state-of-the-art FID of 1.52 on the ImageNet 256x256 benchmark, with a compact generator model of mere 305M parameters.

While Graph Neural Networks (GNNs) are widely used in modern computational biology, an underexplored drawback of common GNN methods,is that they are not inherently multiscale consistent: Two graphs describing the same object or situation at different resolution scales are assigned vastly different latent representations. This prevents graph networks from generating data representations that are consistent across scales. It also complicates the integration of representations at the molecular scale with those generated at the biological scale. Here we discuss why existing GNNs struggle with multiscale consistency and show how to overcome this problem by modifying the message passing paradigm within GNNs.

Model merging combines multiple expert models finetuned from a base foundation model on diverse tasks and domains into a single, more capable model. However, most existing model merging approaches assume that all experts are available simultaneously. In reality, new tasks and domains emerge progressively over time, requiring strategies to integrate the knowledge of expert models as they become available: a process we call temporal model merging. The temporal dimension introduces unique challenges not addressed in prior work, raising new questions such as: when training for a new task, should the expert model start from the merged past experts or from the original base model? Should we merge all models at each time step? Which merging techniques are best suited for temporal merging? Should different strategies be used to initialize the training and deploy the model? To answer these questions, we propose a unified framework called TIME (Temporal Integration of Model Expertise) which defines temporal model merging across three axes: (1) initialization, (2) deployment, and (3) merging technique. Using TIME, we study temporal model merging across model sizes, compute budgets, and learning horizons on the FoMo-in-Flux benchmark. Our comprehensive suite of experiments across TIME allows us to build a better understanding of current challenges and best practices for effective temporal model merging.

Standard graph neural networks assign vastly different latent embeddings to graphs describing the same physical system at different resolution scales. This precludes consistency in applications and prevents generalization between scales as would fundamentally be needed in many scientific applications. We uncover the underlying obstruction, investigate its origin and show how to overcome it.

Traditional SLAM systems, which rely on bundle adjustment, struggle with highly dynamic scenes commonly found in casual videos. Such videos entangle the motion of dynamic elements, undermining the assumption of static environments required by traditional systems. Existing techniques either filter out dynamic elements or model their motion independently. However, the former often results in incomplete reconstructions, whereas the latter can lead to inconsistent motion estimates. Taking a novel approach, this work leverages a 3D point tracker to separate the camera-induced motion from the observed motion of dynamic objects. By considering only the camera-induced component, bundle adjustment can operate reliably on all scene elements as a result. We further ensure depth consistency across video frames with lightweight post-processing based on scale maps. Our framework combines the core of traditional SLAM – bundle adjustment – with a robust learning-based 3D tracker front-end. Integrating motion decomposition, bundle adjustment and depth refinement, our unified framework, BA-Track, accurately tracks the camera motion and produces temporally coherent and scale-consistent dense reconstructions, accommodating both static and dynamic elements. Our experiments on challenging datasets reveal significant improvements in camera pose estimation and 3D reconstruction accuracy.

3D Gaussian Splatting reconstructs scenes by starting from a sparse Structure-from-Motion initialization and iteratively refining under-reconstructed regions. This process is inherently slow, as it requires multiple densification steps where Gaussians are repeatedly split and adjusted, following a lengthy optimization path. Moreover, this incremental approach often leads to suboptimal renderings, particularly in high-frequency regions where detail is critical. We propose a fundamentally different approach: we eliminate densification process with a one-step approximation of scene geometry using triangulated pixels from dense image correspondences. This dense initialization allows us to estimate rough geometry of the scene while preserving rich details from input RGB images, providing each Gaussian with well-informed colors, scales, and positions. As a result, we dramatically shorten the optimization path and remove the need for densification. Unlike traditional methods that rely on sparse keypoints, our dense initialization ensures uniform detail across the scene, even in high-frequency regions where 3DGS and other methods struggle. Moreover, since all splats are initialized in parallel at the start of optimization, we eliminate the need to wait for densification to adjust new Gaussians. Our method not only outperforms speed-optimized models in training efficiency but also achieves higher rendering quality than state-of-the-art approaches, all while using only half the splats of standard 3DGS. It is fully compatible with other 3DGS acceleration techniques, making it a versatile and efficient solution that can be integrated with existing approaches.

Systematic generalization refers to the capacity to understand and generate novel combinations from known components. Despite recent progress by large language models (LLMs) across various domains, these models often fail to extend their knowledge to novel compositional scenarios, revealing notable limitations in systematic generalization. There has been an ongoing debate about whether neural networks possess the capacity for systematic generalization, with recent studies suggesting that meta-learning approaches designed for compositionality can significantly enhance this ability. However, these insights have largely been confined to linguistic problems, leaving their applicability to other tasks an open question. In this study, we extend the approach of meta-learning for compositionality to the domain of abstract spatial reasoning. To this end, we introduce SYGAR-a dataset designed to evaluate the capacity of models to systematically generalize from known geometric transformations (e.g., translation, rotation) of two-dimensional objects to novel combinations of these transformations (e.g., translation+rotation). Our results show that a transformer-based encoder-decoder model, trained via meta-learning for compositionality, can systematically generalize to previously unseen transformation compositions, significantly outperforming state-of-the-art LLMs, including o3-mini, GPT-4o, and Gemini 2.0 Flash, which fail to exhibit similar systematic behavior. Our findings highlight the effectiveness of meta-learning in promoting systematicity beyond linguistic tasks, suggesting a promising direction toward more robust and generalizable models.

Sparse Autoencoders (SAEs) have recently been shown to enhance interpretability and steerability in Large Language Models (LLMs). In this work, we extend the application of SAEs to Vision-Language Models (VLMs), such as CLIP, and introduce a comprehensive framework for evaluating monosemanticity in vision representations. Our experimental results reveal that SAEs trained on VLMs significantly enhance the monosemanticity of individual neurons while also exhibiting hierarchical representations that align well with expert-defined structures (e.g., iNaturalist taxonomy). Most notably, we demonstrate that applying SAEs to intervene on a CLIP vision encoder, directly steer output from multimodal LLMs (e.g., LLaVA) without any modifications to the underlying model. These findings emphasize the practicality and efficacy of SAEs as an unsupervised approach for enhancing both the interpretability and control of VLMs.

Despite the astonishing progress in generative AI, 4D dynamic object generation remains an open challenge. With limited high-quality training data and heavy computing requirements, the combination of hallucinating unseen geometry together with unseen movement poses great challenges to generative models. In this work, we propose TwoSquared as a method to obtain a 4D physically plausible sequence starting from only two 2D RGB images corresponding to the beginning and end of the action. Instead of directly solving the 4D generation problem, TwoSquared decomposes the problem into two steps: 1) an image-to-3D module generation based on the existing generative model trained on high-quality 3D assets, and 2) a physically inspired deformation module to predict intermediate movements. To this end, our method does not require templates or object-class-specific prior knowledge and can take in-the-wild images as input. In our experiments, we demonstrate that TwoSquared is capable of producing texture-consistent and geometry-consistent 4D sequences only given 2D images.

In recent years, text-to-audio systems have achieved remarkable success, enabling the generation of complete audio segments directly from text descriptions. While these systems also facilitate music creation, the element of human creativity and deliberate expression is often limited. In contrast, the present work allows composers, arrangers, and performers to create the basic building blocks for music creation: audio of individual musical notes for use in electronic instruments and DAWs. Through text prompts, the user can specify the timbre characteristics of the audio. We introduce a system that combines a latent diffusion model and multi-modal contrastive learning to generate musical timbres conditioned on text descriptions. By jointly generating the magnitude and phase of the spectrogram, our method eliminates the need for subsequently running a phase retrieval algorithm, as related methods do.

Recent works on the global place recognition treat the task as a retrieval problem, where an off-the-shelf global descriptor is commonly designed in image-based and LiDAR-based modalities. However, it is non-trivial to perform accurate image-LiDAR global place recognition since extracting consistent and robust global descriptors from different domains (2D images and 3D point clouds) is challenging. To address this issue, we propose a novel Voxel-Cross-Pixel (VXP) approach, which establishes voxel and pixel correspondences in a self-supervised manner and brings them into a shared feature space. Specifically, VXP is trained in a two-stage manner that first explicitly exploits local feature correspondences and enforces similarity of global descriptors. Extensive experiments on the three benchmarks (Oxford RobotCar, ViViD++ and KITTI) demonstrate our method surpasses the state-of-the-art cross-modal retrieval by a large margin.

Reconstructing scenes and tracking motion are two sides of the same coin. Tracking points allow for geometric reconstruction [14], while geometric reconstruction of (dynamic) scenes allows for 3D tracking of points over time [24, 39]. The latter was recently also exploited for 2D point tracking to overcome occlusion ambiguities by lifting tracking directly into 3D [38]. However, above approaches either require offline processing or multi-view camera setups both unrealistic for real-world applications like robot navigation or mixed reality. We target the challenge of online 2D and 3D point tracking from unposed monocular camera input introducing Dynamic Online Monocular Reconstruction (DynOMo). We leverage 3D Gaussian splatting to reconstruct dynamic scenes in an online fashion. Our approach extends 3D Gaussians to capture new content and object motions while estimating camera movements from a single RGB frame. DynOMo stands out by enabling emergence of point trajectories through robust image feature reconstruction and a novel similarity-enhanced regularization term, without requiring any correspondence-level supervision. It sets the first baseline for online point tracking with monocular unposed cameras, achieving performance on par with existing methods. We aim to inspire the community to advance online point tracking and reconstruction, expanding the applicability to diverse real-world scenarios.

The interest in matching non-rigidly deformed shapes represented as raw point clouds is rising due to the proliferation of low-cost 3D sensors. Yet, the task is challenging since point clouds are irregular and there is a lack of intrinsic shape information. We propose to tackle these challenges by learning a new shape representation – a per-point high dimensional embedding, in an embedding space where semantically similar points share similar embeddings. The learned embedding has multiple beneficial properties: it is aware of the underlying shape geometry and is robust to shape deformations and various shape artefacts, such as noise and partiality. Consequently, this embedding can be directly employed to retrieve high-quality dense correspondences through a simple nearest neighbor search in the embedding space. Extensive experiments demonstrate new state-of-the-art results and robustness in numerous challenging non-rigid shape matching benchmarks and show its great potential in other shape analysis tasks, such as segmentation.

Objectives: Despite prevalent use of chemical exchange saturation transfer (CEST) MRI, standardization remains elusive. Imaging depends heavily on parameters dictating radiofrequency (RF) events, gradients, and apparent diffusion coefficient (ADC). We present the Pulseq-CEST Library, a repository of CEST preparation and simulation definitions, including example data and evaluations, that provides a common basis for reproducible research, rapid prototyping, and in silico deep learning training data generation.
Materials and methods: A Pulseq-CEST experiment requires (i) a CEST preparation sequence, (ii) a Bloch–McConnell parameter set, (iii) a Bloch–McConnell simulation, and (iv) an evaluation script. Pulseq-CEST utilizes the Bloch–McConnell equations to model in vitro and in vivo conditions. Using this model, a candidate sequence or environment can be held constant while varying other inputs, enabling robust testing.
Results: Data were compared for amide proton transfer weighted (APTw) and water shift and B1 (WASABI) protocols using a five-tube phantom and simulated environments. Real and simulated data matched anticipated spectral shapes and local peak characteristics. The Pulseq-CEST Library supports similar experiments with common sequences and environments to assess new protocols and sample data.
Discussion: The Pulseq-CEST Library provides a flexible mechanism for standardizing and prototyping CEST sequences, facilitating collaborative development. With the capability for expansion, including open-source incorporation of new sequences and environments, the library accelerates the invention and spread of novel CEST and other saturation transfer approaches, such as relayed NOEs (rNOEs) and semisolid magnetization transfer contrast (MTC) methods.

With the increasing use of image generation technology, understanding its social biases, including gender bias, is essential. This paper presents the first large-scale study on gender bias in text-to-image (T2I) models, focusing on everyday situations. While previous research has examined biases in occupations, we extend this analysis to gender associations in daily activities, objects, and contexts. We create a dataset of 3,217 gender-neutral prompts and generate 200 images per prompt from five leading T2I models. We automatically detect the perceived gender of people in the generated images and filter out images with no person or multiple people of different genders, leaving 2,293,295 images. To enable a broad analysis of gender bias in T2I models, we group prompts into semantically similar concepts and calculate the proportion of male- and female-gendered images for each prompt. Our analysis shows that T2I models reinforce traditional gender roles, reflect common gender stereotypes in household roles, and underrepresent women in financial related activities. Women are predominantly portrayed in care- and human-centered scenarios, and men in technical or physical labor scenarios.

We tackle the challenge of LiDAR-based place recognition, which traditionally depends on costly and time-consuming prior 3D maps. To overcome this, we first construct XA-L&RSI dataset, which encompasses approximately 110,000 remote sensing submaps and 13,000 LiDAR point cloud submaps captured in urban scenes, and propose a novel method, L2RSI, for cross-view LiDAR place recognition using high-resolution Remote Sensing Imagery. This approach enables large-scale localization capabilities at a reduced cost by leveraging readily available overhead images as map proxies. L2RSI addresses the dual challenges of cross-view and cross-modal place recognition by learning feature alignment between point cloud submaps and remote sensing submaps in the semantic domain. Additionally, we introduce a novel probability propagation method based on a dynamic Gaussian mixture model to refine position predictions, effectively leveraging temporal and spatial information. This approach enables large-scale retrieval and cross-scene generalization without fine-tuning. Extensive experiments on XA-L&RSI demonstrate that, within a 100km2 retrieval range, L2RSI accurately localizes 95.08% of point cloud submaps within a 30m radius for top-1 retrieved location. We provide a video to more vividly display the place recognition results of L2RSI at this https URL.

Keeping large language models factually up-to-date is crucial for deployment, yet costly retraining remains a challenge. Knowledge editing offers a promising alternative, but methods are only tested on small-scale or synthetic edit benchmarks. In this work, we aim to bridge research into lifelong knowledge editing to real-world edits at practically relevant scale. We first introduce WikiBigEdit; a large-scale benchmark of real-world Wikidata edits, built to automatically extend lifelong for future-proof benchmarking. In its first instance, it includes over 500K question-answer pairs for knowledge editing alongside a comprehensive evaluation pipeline. Finally, we use WikiBigEdit to study existing knowledge editing techniques’ ability to incorporate large volumes of real-world facts and contrast their capabilities to generic modification techniques such as retrieval augmentation and continual finetuning to acquire a complete picture of the practical extent of current lifelong knowledge editing.

Current 3D stylization techniques primarily focus on static scenes, while our world is inherently dynamic, filled with moving objects and changing environments. Existing style transfer methods primarily target appearance – such as color and texture transformation – but often neglect the geometric characteristics of the style image, which are crucial for achieving a complete and coherent stylization effect. To overcome these shortcomings, we propose GAS-NeRF, a novel approach for joint appearance and geometry stylization in dynamic Radiance Fields. Our method leverages depth maps to extract and transfer geometric details into the radiance field, followed by appearance transfer. Experimental results on synthetic and real-world datasets demonstrate that our approach significantly enhances the stylization quality while maintaining temporal coherence in dynamic scenes.

Advances in Earth observation (EO) foundation models have unlocked the potential of big satellite data to learn generic representations from space, benefiting a wide range of downstream applications crucial to our planet. However, most existing efforts remain limited to fixed spectral sensors, focus solely on the Earth’s surface, and overlook valuable metadata beyond imagery. In this work, we take a step towards next-generation EO foundation models with three key components: 1) Copernicus-Pretrain, a massive-scale pretraining dataset that integrates 18.7M aligned images from all major Copernicus Sentinel missions, spanning from the Earth’s surface to its atmosphere; 2) Copernicus-FM, a unified foundation model capable of processing any spectral or non-spectral sensor modality using extended dynamic hypernetworks and flexible metadata encoding; and 3) Copernicus-Bench, a systematic evaluation benchmark with 15 hierarchical downstream tasks ranging from preprocessing to specialized applications for each Sentinel mission. Our dataset, model, and benchmark greatly improve the scalability, versatility, and multimodal adaptability of EO foundation models, while also creating new opportunities to connect EO, weather, and climate research.

Semantic correspondence, the task of determining relationships between different parts of images, underpins various applications including 3D reconstruction, image-to-image translation, object tracking, and visual place recognition. Recent studies have begun to explore representations learned in large generative image models for semantic correspondence, demonstrating promising results. Building on this progress, current state-of-the-art methods rely on combining multiple large models, resulting in high computational demands and reduced efficiency. In this work, we address this challenge by proposing a more computationally efficient approach. We propose a novel knowledge distillation technique to overcome the problem of reduced efficiency. We show how to use two large vision foundation models and distill the capabilities of these complementary models into one smaller model that maintains high accuracy at reduced computational cost. Furthermore, we demonstrate that by incorporating 3D data, we are able to further improve performance, without the need for human-annotated correspondences. Overall, our empirical results demonstrate that our distilled model with 3D data augmentation achieves performance superior to current state-of-the-art methods while significantly reducing computational load and enhancing practicality for real-world applications, such as semantic video correspondence. Our code and weights are publicly available on our project page.

Facade semantic segmentation is a long-standing challenge in photogrammetry and computer vision. Although the last decades have witnessed the influx of facade segmentation methods, there is a lack of comprehensive facade classes and data covering the architectural variability. In ZAHA11Project page: https://github.com/OloOcki/zaha, we introduce Level of Facade Generalization (LoFG), novel hierarchical facade classes designed based on international urban modeling standards, ensuring compatibility with real-world challenging classes and uniform methods’ comparison. Realizing the LoFG, we present to date the largest semantic 3D facade segmentation dataset, providing 601 million annotated points at five and 15 classes of LoFG2 and LoFG3, respectively. More-over, we analyze the performance of baseline semantic segmentation methods on our introduced LoFG classes and data, complementing it with a discussion on the unresolved challenges for facade segmentation. We firmly believe that ZAHA shall facilitate further development of 3D facade semantic segmentation methods, enabling robust segmentation indispensable in creating urban digital twins.

The bidirectional reflectance distribution function (BRDF) is an essential tool to capture the complex interaction of light and matter. Recently, several works have employed neural methods for BRDF modeling, following various strategies, ranging from utilizing existing parametric models to purely neural parametrizations. While all methods yield impressive results, a comprehensive comparison of the different approaches is missing in the literature. In this work, we present a thorough evaluation of several approaches, including results for qualitative and quantitative reconstruction quality and an analysis of reciprocity and energy conservation. Moreover, we propose two extensions that can be added to existing approaches: A novel additive combination strategy for neural BRDFs that split the reflectance into a diffuse and a specular part, and an input mapping that ensures reciprocity exactly by construction, while previous approaches only ensure it by soft constraints.

In this work we propose a novel method for unsupervised controllable video generation. Once trained on a dataset of unannotated videos, at inference our model is capable of both composing scenes of predefined object parts and animating them in a plausible and controlled way. This is achieved by conditioning video generation on a randomly selected subset of local pre-trained self-supervised features during training. We call our model CAGE for visual Composition and Animation for video GEneration. We conduct a series of experiments to demonstrate capabilities of CAGE in various settings.

Accurately describing images with text is a foundation of explainable AI. Vision-Language Models (VLMs) like CLIP have recently addressed this by aligning images and texts in a shared embedding space, expressing semantic similarities between vision and language embeddings. VLM classification can be improved with descriptions generated by Large Language Models (LLMs). However, it is difficult to determine the contribution of actual description semantics, as the performance gain may also stem from a semantic-agnostic ensembling effect, where multiple modified text prompts act as a noisy test-time augmentation for the original one. We propose an alternative evaluation scenario to decide if a performance boost of LLM-generated descriptions is caused by such a noise augmentation effect or rather by genuine description semantics. The proposed scenario avoids noisy test-time augmentation and ensures that genuine, distinctive descriptions cause the performance boost. Furthermore, we propose a training-free method for selecting discriminative descriptions that work independently of classname-ensembling effects. Our approach identifies descriptions that effectively differentiate classes within a local CLIP label neighborhood, improving classification accuracy across seven datasets. Additionally, we provide insights into the explainability of description-based image classification with VLMs.

Current discriminative depth estimation methods often produce blurry artifacts, while generative approaches suffer from slow sampling due to curvatures in the noise-to-depth transport. Our method addresses these challenges by framing depth estimation as a direct transport between image and depth distributions. We are the first to explore flow matching in this field, and we demonstrate that its interpolation trajectories enhance both training and sampling efficiency while preserving high performance. While generative models typically require extensive training data, we mitigate this dependency by integrating external knowledge from a pre-trained image diffusion model, enabling effective transfer even across differing objectives. To further boost our model performance, we employ synthetic data and utilize image-depth pairs generated by a discriminative model on an in-the-wild image dataset. As a generative model, our model can reliably estimate depth confidence, which provides an additional advantage. Our approach achieves competitive zero-shot performance on standard benchmarks of complex natural scenes while improving sampling efficiency and only requiring minimal synthetic data for training.

High-detail semantic 3D building models are frequently utilized in robotics, geoinformatics, and computer vision. One key aspect of creating such models is employing 2D conflict maps that detect openings’ locations in building facades. Yet, in reality, these maps are often incomplete due to obstacles encountered during laser scanning. To address this challenge, we introduce FacaDiffy, a novel method for inpainting unseen facade parts by completing conflict maps with a personalized Stable Diffusion model. Specifically, we first propose a deterministic ray analysis approach to derive 2D conflict maps from existing 3D building models and corresponding laser scanning point clouds. Furthermore, we facilitate the inpainting of unseen facade objects into these 2D conflict maps by leveraging the potential of personalizing a Stable Diffusion model. To complement the scarcity of real-world training data, we also develop a scalable pipeline to produce synthetic conflict maps using random city model generators and annotated facade images. Extensive experiments demonstrate that FacaDiffy achieves state-of-the-art performance in conflict map completion compared to various inpainting baselines and increases the detection rate by 22% when applying the completed conflict maps for high-definition 3D semantic building reconstruction.

Generating long, high-quality videos remains a challenge due to the complex interplay of spatial and temporal dynamics and hardware limitations. In this work, we introduce textbf{MaskFlow}, a unified video generation framework that combines discrete representations with flow-matching to enable efficient generation of high-quality long videos. By leveraging a frame-level masking strategy during training, MaskFlow conditions on previously generated unmasked frames to generate videos with lengths ten times beyond that of the training sequences. MaskFlow does so very efficiently by enabling the use of fast Masked Generative Model (MGM)-style sampling and can be deployed in both fully autoregressive as well as full-sequence generation modes. We validate the quality of our method on the FaceForensics (FFS) and Deepmind Lab (DMLab) datasets and report Fréchet Video Distance (FVD) competitive with state-of-the-art approaches. We also provide a detailed analysis on the sampling efficiency of our method and demonstrate that MaskFlow can be applied to both timestep-dependent and timestep-independent models in a training-free manner.

Understanding neural networks is challenging due to their high-dimensional, interacting components. Inspired by human cognition, which processes complex sensory data by chunking it into recurring entities, we propose leveraging this principle to interpret artificial neural population activities. Biological and artificial intelligence share the challenge of learning from structured, naturalistic data, and we hypothesize that the cognitive mechanism of chunking can provide insights into artificial systems. We first demonstrate this concept in recurrent neural networks (RNNs) trained on artificial sequences with imposed regularities, observing that their hidden states reflect these patterns, which can be extracted as a dictionary of chunks that influence network responses. Extending this to large language models (LLMs) like LLaMA, we identify similar recurring embedding states corresponding to concepts in the input, with perturbations to these states activating or inhibiting the associated concepts. By exploring methods to extract dictionaries of identifiable chunks across neural embeddings of varying complexity, our findings introduce a new framework for interpreting neural networks, framing their population activity as structured reflections of the data they process.

Background: Deep learning-based methods led to significant advancements in many areas of medical imaging, most of which are concerned with the reduction of artifacts caused by motion, scatter, or noise. However, with most neural networks being black boxes, they remain notoriously difficult to interpret, hindering their clinical implementation. In particular, it has been shown that networks exhibit invariances w.r.t. input features, that is, they learn to ignore certain information in the input data.
Purpose: To improve the interpretability of deep learning-based low-dose CT image denoising networks.
Methods: We learn a complete data representation of low-dose input images using a conditional variational autoencoder (cVAE). In this representation, invariances of any given denoising network are then disentangled from the information it is not invariant to using a conditional invertible neural network (cINN). At test time, image-space invariances are generated by applying the inverse of the cINN and subsequent decoding using the cVAE. We propose two methods to analyze sampled invariances and to find those that correspond to alterations of anatomical structures.
Results: The proposed method is applied to four popular deep learning-based low-dose CT image denoising networks. We find that the networks are not only invariant to noise amplitude and realizations, but also to anatomical structures.
Conclusions: The proposed method is capable of reconstructing and analyzing invariances of deep learning-based low-dose CT image denoising networks. This is an important step toward interpreting deep learning-based methods for medical imaging, which is essential for their clinical implementation.

Large language models (LLMs) are being increasingly incorporated into scientific workflows. However, we have yet to fully grasp the implications of this integration. How should the advancement of large language models affect the practice of science? For this opinion piece, we have invited four diverse groups of scientists to reflect on this query, sharing their perspectives and engaging in debate. Schulz et al. make the argument that working with LLMs is not fundamentally different from working with human collaborators, while Bender et al. argue that LLMs are often misused and overhyped, and that their limitations warrant a focus on more specialized, easily interpretable tools. Marelli et al. emphasize the importance of transparent attribution and responsible use of LLMs. Finally, Botvinick and Gershman advocate that humans should retain responsibility for determining the scientific roadmap. To facilitate the discussion, the four perspectives are complemented with a response from each group. By putting these different perspectives in conversation, we aim to bring attention to important considerations within the academic community regarding the adoption of LLMs and their impact on both current and future scientific practices.

Stylizing a dynamic scene based on an exemplar image is critical for various real-world applications, including gaming, filmmaking, and augmented and virtual reality. However, achieving consistent stylization across both spatial and temporal dimensions remains a significant challenge. Most existing methods are designed for static scenes and often require an optimization process for each style image, limiting their adaptability. We introduce ZDySS, a zero-shot stylization framework for dynamic scenes, allowing our model to generalize to previously unseen style images at inference. Our approach employs Gaussian splatting for scene representation, linking each Gaussian to a learned feature vector that renders a feature map for any given view and timestamp. By applying style transfer on the learned feature vectors instead of the rendered feature map, we enhance spatio-temporal consistency across frames. Our method demonstrates superior performance and coherence over state-of-the-art baselines in tests on real-world dynamic scenes, making it a robust solution for practical applications.

We show that variational learning can significantly improve the accuracy and calibration of Low-Rank Adaptation (LoRA) without a substantial increase in the cost. We replace AdamW by the Improved Variational Online Newton (IVON) algorithm to finetune large language models. For Llama-2 with 7 billion parameters, IVON improves the accuracy over AdamW by 2.8% and expected calibration error by 4.6%. The accuracy is also better than the other Bayesian alternatives, yet the cost is lower and the implementation is easier. Our work provides additional evidence for the effectiveness of IVON for large language models.

Text-to-Image (T2I) models have made significant advancements in recent years, but they still struggle to accurately capture intricate details specified in complex compositional prompts. While fine-tuning T2I models with reward objectives has shown promise, it suffers from ‘reward hacking’ and may not generalize well to unseen prompt distributions. In this work, we propose Reward-based Noise Optimization (ReNO), a novel approach that enhances T2I models at inference by optimizing the initial noise based on the signal from one or multiple human preference reward models. Remarkably, solving this optimization problem with gradient ascent for 50 iterations yields impressive results on four different one-step models across two competitive benchmarks, T2I-CompBench and GenEval. Within a computational budget of 20-50 seconds, ReNO-enhanced one-step models consistently surpass the performance of all current open-source Text-to-Image models. Extensive user studies demonstrate that our model is preferred nearly twice as often compared to the popular SDXL model and is on par with the proprietary Stable Diffusion 3 with 8B parameters. Moreover, given the same computational resources, a ReNO-optimized one-step model outperforms widely-used open-source models such as SDXL and PixArt-α, highlighting the efficiency and effectiveness of ReNO in enhancing T2I model performance at inference time.

We introduce the Autoregressive PDE Emulator Benchmark (APEBench), a comprehensive benchmark suite to evaluate autoregressive neural emulators for solving partial differential equations. APEBench is based on JAX and provides a seamlessly integrated differentiable simulation framework employing efficient pseudo-spectral methods, enabling 46 distinct PDEs across 1D, 2D, and 3D. Facilitating systematic analysis and comparison of learned emulators, we propose a novel taxonomy for unrolled training and introduce a unique identifier for PDE dynamics that directly relates to the stability criteria of classical numerical methods. APEBench enables the evaluation of diverse neural architectures, and unlike existing benchmarks, its tight integration of the solver enables support for differentiable physics training and neural-hybrid emulators. Moreover, APEBench emphasizes rollout metrics to understand temporal generalization, providing insights into the long-term behavior of emulating PDE dynamics. In several experiments, we highlight the similarities between neural emulators and numerical simulators.

Multimodal foundation models serve numerous applications at the intersection of vision and language. Still, despite being pretrained on extensive data, they become outdated over time. To keep models updated, research into continual pretraining mainly explores scenarios with either (1) infrequent, indiscriminate updates on large-scale new data, or (2) frequent, sample-level updates. However, practical model deployment often operates in the gap between these two limit cases, as real-world applications often demand adaptation to specific subdomains, tasks or concepts – spread over the entire, varying life cycle of a model. In this work, we complement current perspectives on continual pretraining through a research test bed as well as provide comprehensive guidance for effective continual model updates in such scenarios. We first introduce FoMo-in-Flux, a continual multimodal pretraining benchmark with realistic compute constraints and practical deployment requirements, constructed over 63 datasets with diverse visual and semantic coverage. Using FoMo-in-Flux, we explore the complex landscape of practical continual pretraining through multiple perspectives: (1) A data-centric investigation of data mixtures and stream orderings that emulate real-world deployment situations, (2) a method-centric investigation ranging from simple fine-tuning and traditional continual learning strategies to parameter-efficient updates and model merging, (3) meta learning rate schedules and mechanistic design choices, and (4) the influence of model and compute scaling. Together, our insights provide a practitioner’s guide to continual multimodal pretraining for real-world deployment.

Controllable text-to-image (T2I) diffusion models have shown impressive performance in generating high-quality visual content through the incorporation of various conditions. Current methods, however, exhibit limited performance when guided by skeleton human poses, especially in complex pose conditions such as side or rear perspectives of human figures. To address this issue, we present Stable-Pose, a novel adapter model that introduces a coarse-to-fine attention masking strategy into a vision Transformer (ViT) to gain accurate pose guidance for T2I models. Stable-Pose is designed to adeptly handle pose conditions within pre-trained Stable Diffusion, providing a refined and efficient way of aligning pose representation during image synthesis. We leverage the query-key self-attention mechanism of ViTs to explore the interconnections among different anatomical parts in human pose skeletons. Masked pose images are used to smoothly refine the attention maps based on target pose-related features in a hierarchical manner, transitioning from coarse to fine levels. Additionally, our loss function is formulated to allocate increased emphasis to the pose region, thereby augmenting the model’s precision in capturing intricate pose details. We assessed the performance of Stable-Pose across five public datasets under a wide range of indoor and outdoor human pose scenarios. Stable-Pose achieved an AP score of 57.1 in the LAION-Human dataset, marking around 13% improvement over the established technique ControlNet.

In recent years, black-box machine learning approaches have become a dominant modeling paradigm for knowledge extraction in remote sensing. Despite the potential benefits of uncovering the inner workings of these models with explainable AI, a comprehensive overview summarizing the explainable AI methods used and their objectives, findings, and challenges in remote sensing applications is still missing. In this paper, we address this gap by performing a systematic review to identify the key trends in the field and shed light on novel explainable AI approaches and emerging directions that tackle specific remote sensing challenges. We also reveal the common patterns of explanation interpretation, discuss the extracted scientific insights, and reflect on the approaches used for the evaluation of explainable AI methods. As such, our review provides a complete summary of the state-of-the-art of explainable AI in remote sensing. Further, we give a detailed outlook on the challenges and promising research directions, representing a basis for novel methodological development and a useful starting point for new researchers in the field.

Vision-language models (VLMs), such as CLIP and SigLIP, have found remarkable success in classification, retrieval, and generative tasks. For this, VLMs deterministically map images and text descriptions to a joint latent space in which their similarity is assessed using the cosine similarity. However, a deterministic mapping of inputs fails to capture uncertainties over concepts arising from domain shifts when used in downstream tasks. In this work, we propose post-hoc uncertainty estimation in VLMs that does not require additional training. Our method leverages a Bayesian posterior approximation over the last layers in VLMs and analytically quantifies uncertainties over cosine similarities. We demonstrate its effectiveness for uncertainty quantification and support set selection in active learning. Compared to baselines, we obtain improved and well-calibrated predictive uncertainties, interpretable uncertainty estimates, and sample-efficient active learning. Our results show promise for safety-critical applications of large-scale models.

In generative models, two paradigms have gained attraction in various applications: next-set prediction-based Masked Generative Models and next-noise prediction-based Non-Autoregressive Models, e.g., Diffusion Models. In this work, we propose using discrete-state models to connect them and explore their scalability in the vision domain. First, we conduct a step-by-step analysis in a unified design space across two types of models including timestep-independence, noise schedule, temperature, guidance strength, etc in a scalable manner. Second, we re-cast typical discriminative tasks, e.g., image segmentation, as an unmasking process from [MASK] tokens on a discrete-state model. This enables us to perform various sampling processes, including flexible conditional sampling by only training once to model the joint distribution. All aforementioned explorations lead to our framework named Discrete Interpolants, which enables us to achieve state-of-the-art or competitive performance compared to previous discrete-state based methods in various benchmarks, like ImageNet256, MS COCO, and video dataset FaceForensics. In summary, by leveraging [MASK] in discrete-state models, we can bridge Masked Generative and Non-autoregressive Diffusion models, as well as generative and discriminative tasks.

Recent advances in 3D Gaussian Splatting have shown promising results. Existing methods typically assume static scenes and/or multiple images with prior poses. Dynamics, sparse views, and unknown poses significantly increase the problem complexity due to insufficient geometric constraints. To overcome this challenge, we propose a method that can use only two images without prior poses to fit Gaussians in dynamic environments. To achieve this, we introduce two technical contributions. First, we propose an object-level two-view bundle adjustment. This strategy decomposes dynamic scenes into piece-wise rigid components, and jointly estimates the camera pose and motions of dynamic objects. Second, we design an SE(3) field-driven Gaussian training method. It enables fine-grained motion modeling through learnable per-Gaussian transformations. Our method leads to high-fidelity novel view synthesis of dynamic scenes while accurately preserving temporal consistency and object motion. Experiments on both synthetic and real-world datasets demonstrate that our method significantly outperforms state-of-the-art approaches designed for the cases of static environments, multiple images, and/or known poses.

Vision tokenizers have gained a lot of attraction due to their scalability and compactness; previous works depend on old-school GAN-based hyperparameters, biased comparisons, and a lack of comprehensive analysis of the scaling behaviours. To tackle those issues, we introduce Grouped Spherical Quantization (GSQ), featuring spherical codebook initialization and lookup regularization to constrain codebook latent to a spherical surface. Our empirical analysis of image tokenizer training strategies demonstrates that GSQ-GAN achieves superior reconstruction quality over state-of-the-art methods with fewer training iterations, providing a solid foundation for scaling studies. Building on this, we systematically examine the scaling behaviours of GSQ, specifically in latent dimensionality, codebook size, and compression ratios, and their impact on model performance. Our findings reveal distinct behaviours at high and low spatial compression levels, underscoring challenges in representing high-dimensional latent spaces. We show that GSQ can restructure high-dimensional latent into compact, low-dimensional spaces, thus enabling efficient scaling with improved quality. As a result, GSQ-GAN achieves a 16x down-sampling with a reconstruction FID (rFID) of 0.50.

To date, most place recognition methods focus on single-modality retrieval. While they perform well in specific environments, cross-modal methods offer greater flexibility by allowing seamless switching between map and query sources. It also promises to reduce computation requirements by having a unified model, and achieving greater sample efficiency by sharing parameters. In this work, we develop a universal solution to place recognition, UniLoc, that works with any single query modality (natural language, image, or point cloud). UniLoc leverages recent advances in large-scale contrastive learning, and learns by matching hierarchically at two levels: instance-level matching and scene-level matching. Specifically, we propose a novel Self-Attention based Pooling (SAP) module to evaluate the importance of instance descriptors when aggregated into a place-level descriptor. Experiments on the KITTI-360 dataset demonstrate the benefits of cross-modality for place recognition, achieving superior performance in cross-modal settings and competitive results also for uni-modal scenarios.

We tackle the problem of localizing the traffic surveillance cameras in cooperative perception. To overcome the lack of large-scale real-world intersection datasets, we introduce Carla Intersection, a new simulated dataset with 75 urban and rural intersections in Carla. Moreover, we introduce a novel neural network, TrafficLoc, localizing traffic cameras within a 3D reference map. TrafficLoc employs a coarse-to-fine matching pipeline. For image-point cloud feature fusion, we propose a novel Geometry-guided Attention Loss to address cross-modal viewpoint inconsistencies. During coarse matching, we propose an Inter-Intra Contrastive Learning to achieve precise alignment while preserving distinctiveness among local intra-features within image patch-point group pairs. Besides, we introduce Dense Training Alignment with a soft-argmax operator to consider additional features when regressing the final position. Extensive experiments show that our TrafficLoc improves the localization accuracy over the state-of-the-art Image-to-point cloud registration methods by a large margin (up to 86%) on Carla Intersection and generalizes well to real-world data. TrafficLoc also achieves new SOTA performance on KITTI and NuScenes datasets, demonstrating strong localization ability across both in-vehicle and traffic cameras.

While conventional computer vision emphasizes pixel-level and feature-based objectives, medical image analysis of intricate biological structures necessitates explicit representation of their complex topological properties. Despite their successes, deep learning models often struggle to accurately capture the connectivity and continuity of fine, sometimes pixel-thin, yet critical structures due to their reliance on implicit learning from data. Such shortcomings can significantly impact the reliability of analysis results and hinder clinical decision-making. To address this challenge, we introduce Conformable Convolution, a novel convolutional layer designed to explicitly enforce topological consistency. Conformable Convolution learns adaptive kernel offsets that preferentially focus on regions of high topological significance within an image. This prioritization is guided by our proposed Topological Posterior Generator (TPG) module, which leverages persistent homology. The TPG module identifies key topological features and guides the convolutional layers by applying persistent homology to feature maps transformed into cubical complexes. Our proposed modules are architecture-agnostic, enabling them to be integrated seamlessly into various architectures. We showcase the effectiveness of our framework in the segmentation task, where preserving the interconnectedness of structures is critical. Experimental results on three diverse datasets demonstrate that our framework effectively preserves the topology in the segmentation downstream task, both quantitatively and qualitatively.

Stemming from traditional knowledge graphs (KGs), hyper-relational KGs (HKGs) provide additional key-value pairs (i.e., qualifiers) for each KG fact that help to better restrict the fact validity. In recent years, there has been an increasing interest in studying graph reasoning over HKGs. Meanwhile, as discussed in recent works that focus on temporal KGs (TKGs), world knowledge is ever-evolving, making it important to reason over temporal facts in KGs. Previous mainstream benchmark HKGs do not explicitly specify temporal information for each HKG fact. Therefore, almost all existing HKG reasoning approaches do not devise any module specifically for temporal reasoning. To better study temporal fact reasoning over HKGs, we propose a new type of data structure named hyper-relational TKG (HTKG). Every fact in an HTKG is coupled with a timestamp explicitly indicating its time validity. We develop two new benchmark HTKG datasets, i.e., Wiki-hy and YAGO-hy, and propose an HTKG reasoning model that efficiently models hyper-relational temporal facts. To support future research on this topic, we open-source our datasets and model.

Finding correspondences between 3D shapes is an important and long-standing problem in computer vision, graphics and beyond. While approaches based on machine learning dominate modern 3D shape matching, almost all existing (learning-based) methods require that at least one of the involved shapes is complete. In contrast, the most challenging and arguably most practically relevant setting of matching partially observed shapes, is currently underexplored. One important factor is that existing datasets contain only a small number of shapes (typically below 100), which are unable to serve data-hungry machine learning approaches, particularly in the unsupervised regime. In addition, the type of partiality present in existing datasets is often artificial and far from realistic. To address these limitations and to encourage research on these relevant settings, we provide a generic and flexible framework for the procedural generation of challenging partial shape matching scenarios. Our framework allows for a virtually infinite generation of partial shape matching instances from a finite set of shapes with complete geometry. Further, we manually create cross-dataset correspondences between seven existing (complete geometry) shape matching datasets, leading to a total of 2543 shapes. Based on this, we propose several challenging partial benchmark settings, for which we evaluate respective state-of-the-art methods as baselines.

Understanding dynamic 3D scenes is fundamental for various applications, including extended reality (XR) and autonomous driving. Effectively integrating semantic information into 3D reconstruction enables holistic representation that opens opportunities for immersive and interactive applications. We introduce SADG, Segment Any Dynamic Gaussian Without Object Trackers, a novel approach that combines dynamic Gaussian Splatting representation and semantic information without reliance on object IDs. In contrast to existing works, we do not rely on supervision based on object identities to enable consistent segmentation of dynamic 3D objects. To this end, we propose to learn semantically-aware features by leveraging masks generated from the Segment Anything Model (SAM) and utilizing our novel contrastive learning objective based on hard pixel mining. The learned Gaussian features can be effectively clustered without further post-processing. This enables fast computation for further object-level editing, such as object removal, composition, and style transfer by manipulating the Gaussians in the scene. We further extend several dynamic novel-view datasets with segmentation benchmarks to enable testing of learned feature fields from unseen viewpoints. We evaluate SADG on proposed benchmarks and demonstrate the superior performance of our approach in segmenting objects within dynamic scenes along with its effectiveness for further downstream editing tasks.

We present HI-SLAM2, a geometry-aware Gaussian SLAM system that achieves fast and accurate monocular scene reconstruction using only RGB input. Existing Neural SLAM or 3DGS-based SLAM methods often trade off between rendering quality and geometry accuracy, our research demonstrates that both can be achieved simultaneously with RGB input alone. The key idea of our approach is to enhance the ability for geometry estimation by combining easy-to-obtain monocular priors with learning-based dense SLAM, and then using 3D Gaussian splatting as our core map representation to efficiently model the scene. Upon loop closure, our method ensures on-the-fly global consistency through efficient pose graph bundle adjustment and instant map updates by explicitly deforming the 3D Gaussian units based on anchored keyframe updates. Furthermore, we introduce a grid-based scale alignment strategy to maintain improved scale consistency in prior depths for finer depth details. Through extensive experiments on Replica, ScanNet, and ScanNet++, we demonstrate significant improvements over existing Neural SLAM methods and even surpass RGB-D-based methods in both reconstruction and rendering quality.

Photometric bundle adjustment (PBA) is widely used in estimating the camera pose and 3D geometry by assuming a Lambertian world. However, the assumption of photometric consistency is often violated since the non-diffuse reflection is common in real-world environments. The photometric inconsistency significantly affects the reliability of existing PBA methods. To solve this problem, we propose a novel physically-based PBA method. Specifically, we introduce the physically-based weights regarding material, illumination, and light path. These weights distinguish the pixel pairs with different levels of photometric inconsistency. We also design corresponding models for material estimation based on sequential images and illumination estimation based on point clouds. In addition, we establish the first SLAM-related dataset of non-Lambertian scenes with complete ground truth of illumination and material. Extensive experiments demonstrated that our PBA method outperforms existing approaches in accuracy.

Creating novel views in urban settings is crucial for applications like autonomous driving and virtual tours. Unlike object-level or indoor situations, outdoor settings pose unique challenges, including larger scenes, frame inconsistencies from moving vehicles, and noisy camera poses. This paper introduces a method to address these challenges in view synthesis for outdoor scenarios, utilizing the neural point light field scene representation with 2D image data and 3D point cloud information. We propose a method that efficiently removes dynamic objects in the scene and jointly refines camera poses to recover clean views. We achieve this by estimating the optical flow for the input video sequence and masking out moving objects during training. By learning a consistent geometric representation in the neural point light field, the masked-out areas are correctly recovered in both trained and unseen views, without leaving black areas. Moreover, the learned geometry allows us to extrapolate from current camera trajectory and recover plausible extended views. Additionally, we propose to simultaneously optimize the camera pose along with the scene representation, accommodating noisy camera pose inputs typical of real-world applications. Through validation on real-world urban datasets, we demonstrate stable and satisfactory results in synthesizing novel views of urban scenes.

Historical monuments are a treasure and milestone of cultural heritage. Reconstructing the 3D models of these buildings holds significant value. The rapid development of neural rendering methods makes it possible to recover the original 3D shape exclusively based on archival photographs. However, this task presents considerable challenges due to the properties of available color images. Historical pictures are often limited in number and the scenes in these photos might have altered over time. The radiometric quality of these images is often sub-optimal for using automatic methods. To address these challenges, we introduce an approach to reconstruct the geometry of historical buildings from limited input images. We leverage dense point clouds as a geometric prior and introduce a color appearance embedding loss in volumetric rendering to recover the color of the building. We aim for our work to spark increased interest and focus on preserving historic buildings. Together with the proposed method, we introduce a new historical dataset of the Hungarian National Theater, providing a new benchmark for 3D reconstruction.

Existing techniques for monocular 3D detection have a serious restriction. They tend to perform well only on a limited set of benchmarks, faring well either on ego-centric car views or on traffic camera views, but rarely on both. To encourage progress, this work advocates for an extended evaluation of 3D detection frameworks across different camera perspectives. We make two key contributions. First, we introduce the CARLA Drone dataset, CDrone. Simulating drone views, it substantially expands the diversity of camera perspectives in existing benchmarks. Despite its synthetic nature, CDrone represents a real-world challenge. To show this, we confirm that previous techniques struggle to perform well both on CDrone and a real-world 3D drone dataset. Second, we develop an effective data augmentation pipeline called GroundMix. Its distinguishing element is the use of the ground for creating 3D-consistent augmentation of a training image. GroundMix significantly boosts the detection accuracy of a lightweight one-stage detector. In our expanded evaluation, we achieve the average precision on par with or substantially higher than the previous state of the art across all tested datasets.

Neural implicit representations have become a popular choice for modeling surfaces due to their adaptability in resolution and support for complex topology. While previous works have achieved impressive reconstruction quality by training on ground truth point clouds or meshes, they often do not discuss the data acquisition and ignore the effect of input quality and sampling methods during reconstruction. In this paper, we introduce a method that directly digests depth images for the task of high-fidelity 3D reconstruction. To this end, a novel local geometry feature computation method is proposed such that a simple sampling strategy can be adopted to generate highly effective training data. Due to its simplicity, our sampling strategy can be easily incorporated into diverse popular methods, allowing their training process to be more stable and efficient. Despite its simplicity, our method outperforms a range of both classical and learning-based baselines and demonstrates state-of-the-art results in both synthetic and real-world datasets.

3D scene stylization extends the work of neural style transfer to 3D. A vital challenge in this problem is to maintain the uniformity of the stylized appearance across multiple views. A vast majority of the previous works achieve this by training a 3D model for every stylized image and a set of multi-view images. In contrast, we propose a novel architecture trained on a collection of style images that, at test time, produces real time high-quality stylized novel views. We choose the underlying 3D scene representation for our model as 3D Gaussian splatting. We take the 3D Gaussians and process them using a multi-resolution hash grid and a tiny MLP to obtain stylized views. The MLP is conditioned on different style codes for generalization to different styles during test time. The explicit nature of 3D Gaussians gives us inherent advantages over NeRF-based methods, including geometric consistency and a fast training and rendering regime. This enables our method to be useful for various practical use cases, such as augmented or virtual reality. We demonstrate that our method achieves state-of-the-art performance with superior visual quality on various indoor and outdoor real-world data.

Direct Preference Optimization (DPO) has emerged as a powerful approach to align text-to-image (T2I) models with human feedback. Unfortunately, successful application of DPO to T2I models requires a huge amount of resources to collect and label large-scale datasets, e.g., millions of generated paired images annotated with human preferences. In addition, these human preference datasets can get outdated quickly as the rapid improvements of T2I models lead to higher quality images. In this work, we investigate a scalable approach for collecting large-scale and fully synthetic datasets for DPO training. Specifically, the preferences for paired images are generated using a pre-trained reward function, eliminating the need for involving humans in the annotation process, greatly improving the dataset collection efficiency. Moreover, we demonstrate that such datasets allow averaging predictions across multiple models and collecting ranked preferences as opposed to pairwise preferences. Furthermore, we introduce RankDPO to enhance DPO-based methods using the ranking feedback. Applying RankDPO on SDXL and SD3-Medium models with our synthetically generated preference dataset ‘Syn-Pic’ improves both prompt-following (on benchmarks like T2I-Compbench, GenEval, and DPG-Bench) and visual quality (through user studies). This pipeline presents a practical and scalable solution to develop better preference datasets to enhance the performance of text-to-image models.

In frame-based vision, object detection faces substantial performance degradation under challenging conditions due to the limited sensing capability of conventional cameras. Event cameras output sparse and asynchronous events, providing a potential solution to solve these problems. However, effectively fusing two heterogeneous modalities remains an open issue. In this work, we propose a novel hierarchical feature refinement network for event-frame fusion. The core concept is the design of the coarse-to-fine fusion module, denoted as the cross-modality adaptive feature refinement (CAFR) module. In the initial phase, the bidirectional cross-modality interaction (BCI) part facilitates information bridging from two distinct sources. Subsequently, the features are further refined by aligning the channel-level mean and variance in the two-fold adaptive feature refinement (TAFR) part. We conducted extensive experiments on two benchmarks: the low-resolution PKU-DDD17-Car dataset and the high-resolution DSEC dataset. Experimental results show that our method surpasses the state-of-the-art by an impressive margin of 8% on the DSEC dataset. Besides, our method exhibits significantly better robustness (69.5% versus 38.7%) when introducing 15 different corruption types to the frame images.

Spherical or omni-directional images offer an immersive visual format appealing to a wide range of computer vision applications. However, geometric properties of spherical images pose a major challenge for models and metrics designed for ordinary 2D images. Here, we show that direct application of Fréchet Inception Distance (FID) is insufficient for quantifying geometric fidelity in spherical images. We introduce two quantitative metrics accounting for geometric constraints, namely Omnidirectional FID (OmniFID) and Discontinuity Score (DS). OmniFID is an extension of FID tailored to additionally capture field-of-view requirements of the spherical format by leveraging cubemap projections. DS is a kernel-based seam alignment score of continuity across borders of 2D representations of spherical images. In experiments, OmniFID and DS quantify geometry fidelity issues that are undetected by FID.

Neural implicit surfaces can be used to recover accurate 3D geometry from imperfect point clouds. In this work, we show that state-of-the-art techniques work by minimizing an approximation of a one-sided Chamfer distance. This shape metric is not symmetric, as it only ensures that the point cloud is near the surface but not vice versa. As a consequence, existing methods can produce inaccurate reconstructions with spurious surfaces. Although one approach against spurious surfaces has been widely used in the literature, we theoretically and experimentally show that it is equivalent to regularizing the surface area, resulting in over-smoothing. As a more appealing alternative, we propose DiffCD, a novel loss function corresponding to the symmetric Chamfer distance. In contrast to previous work, DiffCD also assures that the surface is near the point cloud, which eliminates spurious surfaces without the need for additional regularization. We experimentally show that DiffCD reliably recovers a high degree of shape detail, substantially outperforming existing work across varying surface complexity and noise levels.

The diffusion model has long been plagued by scalability and quadratic complexity issues, especially within transformer-based structures. In this study, we aim to leverage the long sequence modeling capability of a State-Space Model called Mamba to extend its applicability to visual data generation. Firstly, we identify a critical oversight in most current Mamba-based vision methods, namely the lack of consideration for spatial continuity in the scan scheme of Mamba. Secondly, building upon this insight, we introduce Zigzag Mamba, a simple, plug-and-play, minimal-parameter burden, DiT style solution, which outperforms Mamba-based baselines and demonstrates improved speed and memory utilization compared to transformer-based baselines, also this heterogeneous layerwise scan enables zero memory and speed burden when we consider more scan paths. Lastly, we integrate Zigzag Mamba with the Stochastic Interpolant framework to investigate the scalability of the model on large-resolution visual datasets, such as FacesHQ and UCF101, MultiModal-CelebA-HQ, and MS COCO .

In Composed Video Retrieval, a video and a textual description which modifies the video content are provided as inputs to the model. The aim is to retrieve the relevant video with the modified content from a database of videos. In this challenging task, the first step is to acquire large-scale training datasets and collect high-quality benchmarks for evaluation. In this work, we introduce EgoCVR, a new evaluation benchmark for fine-grained Composed Video Retrieval using large-scale egocentric video datasets. EgoCVR consists of 2,295 queries that specifically focus on high-quality temporal video understanding. We find that existing Composed Video Retrieval frameworks do not achieve the necessary high-quality temporal video understanding for this task. To address this shortcoming, we adapt a simple training-free method, propose a generic re-ranking framework for Composed Video Retrieval, and demonstrate that this achieves strong results on EgoCVR.

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby hindering the generalization of classification models trained on synthetic datasets. We propose DataDream, a framework for synthesizing classification datasets that more faithfully represents the real data distribution when guided by few-shot examples of the target classes. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets. We demonstrate the efficacy of DataDream through extensive experiments, surpassing state-of-the-art classification accuracy with few-shot data across 7 out of 10 datasets, while being competitive on the other 3. Additionally, we provide insights into the impact of various factors, such as the number of real-shot and generated images as well as the fine-tuning compute on model performance.

While style transfer techniques have been well-developed for 2D image stylization, the extension of these methods to 3D scenes remains relatively unexplored. Existing approaches demonstrate proficiency in transferring colors and textures but often struggle with replicating the geometry of the scenes. In our work, we leverage an explicit Gaussian Scale (GS) representation and directly match the distributions of Gaussians between style and content scenes using the Earth Mover’s Distance (EMD). By employing the entropy-regularized Wasserstein-2 distance, we ensure that the transformation maintains spatial smoothness. Additionally, we decompose the scene stylization problem into smaller chunks to enhance efficiency. This paradigm shift reframes stylization from a pure generative process driven by latent space losses to an explicit matching of distributions between two Gaussian representations. Our method achieves high-resolution 3D stylization by faithfully transferring details from 3D style scenes onto the content scene. Furthermore, WaSt-3D consistently delivers results across diverse content and style scenes without necessitating any training, as it relies solely on optimization-based techniques.

Plane adjustment (PA) is crucial for many 3D applications, involving simultaneous pose estimation and plane recovery. Despite recent advancements, it remains a challenging problem in the realm of multi-view point cloud registration. Current state-of-the-art methods can achieve globally optimal convergence only with good initialization. Furthermore, their high time complexity renders them impractical for large-scale problems. To address these challenges, we first exploit a novel optimization strategy termed Bi-Convex Relaxation, which decouples the original problem into two simpler sub-problems, reformulates each sub-problem using a convex relaxation technique, and alternately solves each one until the original problem converges. Building on this strategy, we propose two algorithmic variants for solving the plane adjustment problem, namely GlobalPointer and GlobalPointer++, based on point-to-plane and plane-to-plane errors, respectively. Extensive experiments on both synthetic and real datasets demonstrate that our method can perform large-scale plane adjustment with linear time complexity, larger convergence region, and robustness to poor initialization, while achieving similar accuracy as prior methods.

Parametric feature grid encodings have gained significant attention as an encoding approach for neural fields since they allow for much smaller MLPs, which significantly decreases the inference time of the models. In this work, we propose MeshFeat, a parametric feature encoding tailored to meshes, for which we adapt the idea of multi-resolution feature grids from Euclidean space. We start from the structure provided by the given vertex topology and use a mesh simplification algorithm to construct a multi-resolution feature representation directly on the mesh. The approach allows the usage of small MLPs for neural fields on meshes, and we show a significant speed-up compared to previous representations while maintaining comparable reconstruction quality for texture reconstruction and BRDF representation. Given its intrinsic coupling to the vertices, the method is particularly well-suited for representations on deforming meshes, making it a good fit for object animation.

Text-to-image generative models have become a prominent and powerful tool that excels at generating high-resolution realistic images. However, guiding the generative process of these models to take into account detailed forms of conditioning reflecting style and/or structure information remains an open problem. In this paper, we present. LoRAdapter, an approach that unifies both style and structure conditioning under the same formulation using a novel conditional LoRA block that enables zero-shot control. LoRAdapter is an efficient and powerful approach to condition text-to-image diffusion models, which enables fine-grained control conditioning during generation and outperforms recent state-of-the-art approaches.

Most Bundle Adjustment (BA) solvers like the Levenberg-Marquardt algorithm require a good initialization. Instead, initialization-free BA remains a largely uncharted territory. The under-explored Variable Projection algorithm (VarPro) exhibits a wide convergence basin even without initialization. Coupled with object space error formulation, recent works have shown its ability to solve small-scale initialization-free bundle adjustment problem. To make such initialization-free BA approaches scalable, we introduce Power Variable Projection (PoVar), extending a recent inverse expansion method based on power series. Importantly, we link the power series expansion to Riemannian manifold optimization. This projective framework is crucial to solve large-scale bundle adjustment problems without initialization. Using the real-world BAL dataset, we experimentally demonstrate that our solver achieves state-of-the-art results in terms of speed and accuracy. To our knowledge, this work is the first to address the scalability of BA without initialization opening new venues for initialization-free structure-from-motion.

3D single object tracking (SOT) is an essential task in autonomous driving and robotics. However, learning robust 3D SOT trackers remains challenging due to the limited category-specific point cloud data and the inherent sparsity and incompleteness of LiDAR scans. To tackle these issues, we propose a unified 3D SOT framework that leverages 3D generative pre-training and learns robust 3D matching abilities from 2D pre-trained foundation trackers. Our framework features a consistent target-matching architecture with the widely used 2D trackers, facilitating the transfer of 2D matching knowledge. Specifically, we first propose a lightweight Target-Aware Projection (TAP) module, allowing the pre-trained 2D tracker to work well on the projected point clouds without further fine-tuning. We then propose a novel IoU-guided matching-distillation framework that utilizes the powerful 2D pre-trained trackers to guide 3D matching learning in the 3D tracker, i.e., the 3D template-to-search matching should be consistent with its corresponding 2D template-to-search matching obtained from 2D pre-trained trackers. Our designs are applied to two mainstream 3D SOT frameworks: memory-less Siamese and contextual memory-based approaches, which are respectively named SiamDisst and MemDisst. Extensive experiments show that SiamDisst and MemDisst achieve state-of-the-art performance on KITTI, Waymo Open Dataset and nuScenes benchmarks, while running at above real-time speed of 25 and 90 FPS on a RTX3090 GPU.

Unsupervised Domain Adaptation (UDA) is the task of bridging the domain gap between a labeled source domain, e.g., synthetic data, and an unlabeled target domain. We observe that current UDA methods show inferior results on fine structures and tend to oversegment objects with ambiguous appearance. To address these shortcomings, we propose to leverage geometric information, i.e., depth predictions, as depth discontinuities often coincide with segmentation boundaries. We show that naively incorporating depth into current UDA methods does not fully exploit the potential of this complementary information. To this end, we present MICDrop, which learns a joint feature representation by masking image encoder features while inversely masking depth encoder features. With this simple yet effective complementary masking strategy, we enforce the use of both modalities when learning the joint feature representation. To aid this process, we propose a feature fusion module to improve both global as well as local information sharing while being robust to errors in the depth predictions. We show that our method can be plugged into various recent UDA methods and consistently improve results across standard UDA benchmarks, obtaining new state-of-the-art performances.

Visual synthesis has recently seen significant leaps in performance, largely due to breakthroughs in generative models. Diffusion models have been a key enabler, as they excel in image diversity. However, this comes at the cost of slow training and synthesis, which is only partially alleviated by latent diffusion. To this end, flow matching is an appealing approach due to its complementary characteristics of faster training and inference but less diverse synthesis. We demonstrate our FMBoost approach, which introduces flow matching between a frozen diffusion model and a convolutional decoder that enables high-resolution image synthesis at reduced computational cost and model size. A small diffusion model can then effectively provide the necessary visual diversity, while flow matching efficiently enhances resolution and detail by mapping the small to a high-dimensional latent space, producing high-resolution images. Combining the diversity of diffusion models, the efficiency of flow matching, and the effectiveness of convolutional decoders, state-of-the-art high-resolution image synthesis is achieved at 10242 pixels with minimal computational cost. Cascading FMBoost optionally boosts this further to 20482 pixels. Importantly, this approach is orthogonal to recent approximation and speed-up strategies for the underlying model, making it easily integrable into the various diffusion model frameworks.

Historical monuments are a treasure and milestone of cultural heritage. Reconstructing the 3D models of these buildings holds significant value. The rapid development of neural rendering methods makes it possible to recover the original 3D shape exclusively based on archival photographs. However, this task presents considerable challenges due to the properties of available color images. Historical pictures are often limited in number and the scenes in these photos might have altered over time. The radiometric quality of these images is often sub-optimal for using automatic methods. To address these challenges, we introduce an approach to reconstruct the geometry of historical buildings from limited input images. We leverage dense point clouds as a geometric prior and introduce a color appearance embedding loss in volumetric rendering to recover the color of the building. We aim for our work to spark increased interest and focus on preserving historic buildings. Together with the proposed method, we introduce a new historical dataset of the Hungarian National Theater, providing a new benchmark for 3D reconstruction.

Neural implicits are a widely used surface presentation because they offer an adaptive resolution and support arbitrary topology changes. While previous works rely on ground truth point clouds or meshes, they often do not discuss the data acquisition and ignore the effect of input quality and sampling methods during reconstruction. In this paper, we introduce a sampling method with an uncertainty-augmented surface implicit representation that employs a sampling technique that considers the geometric characteristics of inputs. To this end, we introduce a strategy that efficiently computes differentiable geometric features, namely, mean curvatures, to guide the sampling phase during the training period. The uncertainty augmentation offers insights into the occupancy and reliability of the output signed distance value, thereby expanding representation capabilities into open surfaces. Finally, we demonstrate that our method improves the reconstruction of both synthetic and real-world data.