Reliable Uncertainty Quantification (UQ) and failure prediction remain open challenges for Vision-Language Models (VLMs). We introduce ViLU, a new Vision-Language Uncertainty quantification framework that contextualizes uncertainty estimates by leveraging all task-relevant textual representations. ViLU constructs an uncertainty-aware multi-modal representation by integrating the visual embedding, the predicted textual embedding, and an image-conditioned textual representation via cross-attention. Unlike traditional UQ methods based on loss prediction, ViLU trains an uncertainty predictor as a binary classifier to distinguish correct from incorrect predictions using a weighted binary cross-entropy loss, making it loss-agnostic. In particular, our proposed approach is well-suited for post-hoc settings, where only vision and text embeddings are available without direct access to the model itself. Extensive experiments on diverse datasets show the significant gains of our method compared to state-of-the-art failure prediction methods. We apply our method to standard classification datasets, such as ImageNet-1k, as well as large-scale image-caption datasets like CC12M and LAION-400M. Ablation studies highlight the critical role of our architecture and training in achieving effective uncertainty quantification.
论文检索
输入标题、作者或关键词，从 8,216 篇学术成果中精准定位
Visual prompt tuning (VPT), i.e., fine-tuning some lightweight prompt tokens, provides an efficient and effective approach for adapting pre-trained models to various downstream tasks. However, most prior art indiscriminately uses a fixed prompt distribution across different tasks, neglecting the importance of each block varying depending on the task. In this paper, we introduce adaptive distribution optimization (ADO) by tackling two key questions: (1) How to appropriately and formally define ADO, and (2) How to design an adaptive distribution strategy guided by this definition? Through empirical analysis, we first confirm that properly adjusting the distribution significantly improves VPT performance, and further uncover a key insight that a nested relationship exists between ADO and VPT. Based on these findings, we propose a new VPT framework, termed PRO-VPT (iterative Prompt RelOcation-based VPT), which adaptively adjusts the distribution built upon a nested optimization formulation. Specifically, we develop a prompt relocation strategy derived from this formulation, comprising two steps: pruning idle prompts from prompt-saturated blocks, followed by allocating these prompts to the most prompt-needed blocks. By iteratively performing prompt relocation and VPT, our proposal can adaptively learn the optimal prompt distribution in a nested optimization-based manner, thereby unlocking the full potential of VPT. Extensive experiments demonstrate that our proposal significantly outperforms advanced VPT methods, e.g., PRO-VPT surpasses VPT by 1.6 pp and 2.0 pp average accuracy, leading prompt-based methods to state-of-the-art performance on VTAB-1k and FGVC benchmarks. The code is available at https://github.com/ckshang/PRO-VPT.
Capturing accurate 3D human pose in the wild would provide valuable data for training motion-generation and pose-estimation methods. While video-based estimation approaches have become increasingly accurate, they often fail in common scenarios involving self-contact, such as a hand touching the face. In contrast, wearable bioimpedance sensing can cheaply and unobtrusively measure ground-truth skin-to-skin contact. Consequently, we propose a novel framework that combines visual pose estimators with bioimpedance sensing to capture the 3D pose of people by taking self-contact into account. Our method, BioTUCH, initializes the pose using an off-the-shelf estimator and introduces contact-aware pose optimization during measured self-contact: reprojection error and deviations from the input estimate are minimized while enforcing vertex proximity constraints. We validate our approach using a new dataset of synchronized RGB video, bioimpedance measurements, and 3D motion capture. Testing with three input pose estimators, we demonstrate an average of 11.7% improvement in reconstruction accuracy. We also present a miniature wearable bioimpedance sensor that enables efficient large-scale collection of contact-aware training data for improving pose estimation and generation using BioTUCH. Code and data are available at biotuch.is.tue.mpg.de
InsideOut: Integrated RGB-Radiative Gaussian Splatting for Comprehensive 3D Object Representation
PDF ↗We introduce InsideOut, an extension of 3D Gaussian splatting (3DGS) that bridges the gap between high-fidelity RGB surface details and subsurface X-ray structures. The fusion of RGB and X-ray imaging is invaluable in fields such as medical diagnostics, cultural heritage restoration, and manufacturing. We collect new paired RGB and X-ray data, perform hierarchical fitting to align RGB and X-ray radiative Gaussian splats, and propose an X-ray reference loss to ensure consistent internal structures. InsideOut effectively addresses the challenges posed by disparate data representations between the two modalities and limited paired datasets. This approach significantly extends the applicability of 3DGS, enhancing visualization, simulation, and non-destructive testing capabilities across various domains.
Co-speech gestures play a vital role in non-verbal communication. In this paper, we introduce a new framework for co-speech gesture understanding in the wild. Specifically, we propose three new tasks and benchmarks to evaluate a model's capability to comprehend gesture-speech-text associations: (i) gesture based retrieval, (ii) gestured word spotting, and (iii) active speaker detection using gestures. We present a new approach that learns a tri-modal video-gesture-speech-text representation to solve these tasks. By leveraging a combination of global phrase contrastive loss and local gesture-word coupling loss, we demonstrate that a strong gesture representation can be learned in a weakly supervised manner from videos in the wild. Our learned representations outperform previous methods, including large vision-language models (VLMs). Further analysis reveals that speech and text modalities capture distinct gesture related signals, underscoring the advantages of learning a shared tri-modal embedding space. The dataset, model, and code are available at: https://www.robots.ox.ac.uk/ vgg/research/jegal.
When Large Vision-Language Model Meets Large Remote Sensing Imagery: Coarse-to-Fine Text-Guided Token Pruning
PDF ↗Efficient vision-language understanding of large Remote Sensing Images (RSIs) is meaningful but challenging. Current Large Vision-Language Models (LVLMs) typically employ limited pre-defined grids to process images, leading to information loss when handling gigapixel RSIs. Conversely, using unlimited grids significantly increases computational costs. To preserve image details while reducing computational complexity, we propose a text-guided token pruning method with Dynamic Image Pyramid (DIP) integration. Our method introduces: (i) a Region Focus Module (RFM) that leverages text-aware region localization capability to identify critical vision tokens, and (ii) a coarse-to-fine image tile selection and vision token pruning strategy based on DIP, which is guided by RFM outputs and avoids directly processing the entire large imagery. Additionally, existing benchmarks for evaluating LVLMs' perception ability on large RSI suffer from limited question diversity and constrained image sizes. We construct a new benchmark named LRS-VQA, which contains 7,333 QA pairs across 8 categories, with image length up to 27,328 pixels. Our method outperforms existing high-resolution strategies on four datasets using the same data. Moreover, compared to existing token reduction methods, our approach demonstrates higher efficiency under high-resolution settings.
Standard convolutions are prevalent in image processing and deep learning, but their fixed kernels limits adaptability. Several deformation strategies of the reference kernel grid have been proposed. Yet, they lack a unified theoretical framework. By returning to a metric perspective for images, now seen as two-dimensional manifolds equipped with notions of local and geodesic distances, either symmetric (Riemannian) or not (Finsler), we provide a unifying principle: the kernel positions are samples of unit balls of implicit metrics. With this new perspective, we also propose metric convolutions, a novel approach that samples unit balls from explicit signal-dependent metrics, providing interpretable operators with geometric regularisation. This framework, compatible with gradient-based optimisation, can directly replace existing convolutions applied to either input images or deep features of neural networks. Metric convolutions typically require fewer parameters and provide better generalisation. Our approach shows competitive performance in standard denoising and classification tasks.
Accurate, real-time 3D reconstruction of human heads from monocular images and videos underlies numerous visual applications. As 3D ground truth data is hard to come by at scale, previous methods have sought to learn from abundant 2D videos in a self-supervised manner. Typically, this involves the use of differentiable mesh rendering, which is effective but faces limitations. To improve on this, we propose SHeaP (Self-supervised Head Geometry Predictor Learned via 2D Gaussians).Given a source image, we predict a 3DMM mesh and a set of Gaussians that are rigged to this mesh. We then reanimate this rigged head avatar to match a target frame, and backpropagate photometric losses to both the 3DMM and Gaussian prediction networks. We find that using Gaussians for rendering substantially improves the effectiveness of this self-supervised approach.Training solely on 2D data, our method surpasses existing self-supervised approaches in geometric evaluations on the NoW benchmark for neutral faces and a new benchmark for non-neutral expressions. Our method also produces highly expressive meshes, outperforming state-of-the-art in emotion classification.
Visual object tracking has gained promising progress in past decades. Most of the existing approaches focus on learning target representation in well-conditioned daytime data, while for the unconstrained real-world scenarios with adverse weather conditions, e.g. nighttime or foggy environment, the tremendous domain shift leads to significant performance degradation. In this paper, we propose UMDATrack, which is capable of maintaining high-quality target state prediction under various adverse weather conditions within a unified domain adaptation framework. Specifically, we first use a controllable scenario generator to synthesize a small amount of unlabeled videos (less than 2% frames in source daytime datasets) in multiple weather conditions under the guidance of different text prompts. Afterwards, we design a simple yet effective domain-customized adapter (DCA), allowing the target objects' representation to rapidly adapt to various weather conditions without redundant model updating. Furthermore, to enhance the localization consistency between source and target domains, we propose a target-aware confidence alignment module (TCA) following optimal transport theorem. Extensive experiments demonstrate that UMDATrack can surpass existing advanced visual trackers and lead new state-of-the-art performance by a significant margin.
Structuring latent representations in a hierarchical manner enables models to learn patterns at multiple levels of abstraction. However, most prevalent image understanding models focus on visual similarity, and learning visual hierarchies is relatively unexplored. In this work, for the first time, we introduce a learning paradigm that can encode user-defined multi-level complex visual hierarchies in hyperbolic space without requiring explicit hierarchical labels. As a concrete example, first, we define a part-based image hierarchy using object-level annotations within and across images. Then, we introduce an approach to enforce the hierarchy using contrastive loss with pairwise entailment metrics. Finally, we discuss new evaluation metrics to effectively measure hierarchical image retrieval. Encoding these complex relationships ensures that the learned representations capture semantic and structural information that transcends mere visual similarity. Experiments in part-based image retrieval show significant improvements in hierarchical retrieval tasks, demonstrating the capability of our model in capturing visual hierarchies.
We present Magic Insert, a method to drag-and-drop subjects from a user-provided image into a target image of a different style in a plausible manner while matching the style of the target image. This work formalizes our version of the problem of style-aware drag-and-drop and proposes to tackle it by decomposing it into two sub-problems: style-aware personalization and realistic object insertion in stylized images. For style-aware personalization, we cast our method as a weight-and-text-embedding finetuning method with inference-time module-targeted style injection. For subject insertion, we propose Bootstrapped Domain Adaption (BDA) to adapt a domain-specific photorealistic object insertion model to the domain of diverse artistic styles. Overall, the method significantly outperforms traditional and state-of-the-art approaches that struggle with quality, subject fidelity and harmonious stylization. Finally, we present a new dataset, SubjectPlop, to facilitate evaluation and future progress in this area.
We present DPoser-X, a diffusion-based prior model for 3D whole-body human poses. Building a versatile and robust full-body human pose prior remains challenging due to the inherent complexity of articulated human poses and the scarcity of high-quality whole-body pose datasets. To address these limitations, we introduce a Diffusion model as body Pose prior (DPoser) and extend it to DPoser-X for expressive whole-body human pose modeling.Our approach unifies various pose-centric tasks as inverse problems, solving them through variational diffusion sampling. To enhance performance on downstream applications, we introduce a novel truncated timestep scheduling method specifically designed for pose data characteristics. We also propose a masked training mechanism that effectively combines whole-body and part-specific datasets, enabling our model to capture interdependencies between body parts while avoiding overfitting to specific actions.Extensive experiments demonstrate DPoser-X's robustness and versatility across multiple benchmarks for body, hand, face, and full-body pose modeling. Our model consistently outperforms state-of-the-art alternatives, establishing a new benchmark for whole-body human pose prior modeling.
Category-agnostic pose estimation (CAPE) has traditionally relied on support images with annotated keypoints, a process that is often cumbersome and may fail to fully capture the necessary correspondences across diverse object categories. Recent efforts have explored the use of text queries, leveraging their enhanced stability and generalization capabilities. However, existing approaches often remain constrained by their reliance on support queries, their failure to fully utilize the rich priors embedded in pre-trained large language models, and the limitations imposed by their parametric distribution assumptions. To address these challenges, we introduce CapeLLM, the first multimodal large language model (MLLM) designed for CAPE. Our method only employs query image and detailed text descriptions as an input to estimate category-agnostic keypoints. Our method encompasses effective training strategies and carefully designed instructions for applying the MLLM to CAPE. Moreover, we propose an inference mechanism that further enhances the reasoning process for unseen keypoints. while flexibly modeling their underlying spatial distribution and uncertainty, allowing for adaptive refinement based on contextual cues. We conducted extensive experiments to apply the MLLM to CAPE effectively, focusing not only on the model architecture and prompt design but also on ensuring robustness across input variations. Our approach sets a new state-of-the-art on the MP-100 benchmark in the 1-shot and even 5-shot setting, marking a significant advancement in the field of category-agnostic pose estimation. Code is available https://github.com/Junhojuno/CapeLLM
In this paper, we present DM-Calib, a diffusion-based approach for estimating pinhole camera intrinsic parameters from a single input image. Monocular camera calibration is essential for many 3D vision tasks. However, most existing methods depend on handcrafted assumptions or are constrained by limited training data, resulting in poor generalization across diverse real-world images. Recent advancements in stable diffusion models, trained on massive data, have shown the ability to generate high-quality images with varied characteristics. Emerging evidence indicates that these models implicitly capture the relationship between camera focal length and image content. Building on this insight, we explore how to leverage the powerful priors of diffusion models for monocular pinhole camera calibration. Specifically, we introduce a new image-based representation, termed Camera Image, which losslessly encodes the numerical camera intrinsics and integrates seamlessly with the diffusion framework. Using this representation, we reformulate the problem of estimating camera intrinsics as the generation of a dense Camera Image conditioned on an input image. By fine-tuning a stable diffusion model to generate a Camera Image from a single RGB input, we can extract camera intrinsics via a RANSAC operation. We further demonstrate that our monocular calibration method enhances performance across various 3D tasks, including zero-shot metric depth estimation, 3D metrology, pose estimation and sparse-view reconstruction. Extensive experiments on multiple public datasets show that our approach significantly outperforms baselines and provides broad benefits to 3D vision tasks.
Generating human videos from a single image while ensuring high visual quality and precise control is a challenging task, especially in complex scenarios involving multiple individuals and interactions with objects. Existing methods, while effective for single-human cases, often fail to handle the intricacies of multi-identity interactions because they struggle to associate the correct pairs of human appearance and pose condition and model the distribution of 3D-aware dynamics. To address these limitations, we present Structural Video Diffusion, a novel framework designed for generating realistic multi-human videos. Our approach introduces two core innovations: identity-specific embeddings to maintain consistent appearances across individuals and a structural learning mechanism that incorporates depth and surface-normal cues to model human-object interactions. Additionally, we expand existing human video dataset with 25K new videos featuring diverse multi-human and object interaction scenarios, providing a robust foundation for training. Experimental results demonstrate that Structural Video Diffusion achieves superior performance in generating lifelike, coherent videos for multiple subjects with dynamic and rich interactions, advancing the state of human-centric video generation. Code is available at https://github.com/zhenzhiwang/Multi-HumanVid
Although autonomous driving systems demonstrate high perception performance, they still face limitations when handling rare situations or complex road structures. Since existing road infrastructures are designed for human drivers, safety improvements are typically introduced only after accidents occur. This reactive approach poses a significant challenge for autonomous systems, which require proactive risk mitigation. To address this issue, we propose OD-RASE, a framework for enhancing the safety of autonomous driving systems by detecting road structures that cause traffic accidents and connecting these findings to infrastructure development. First, we formalize an ontology based on specialized domain knowledge of road traffic systems. In parallel, we generate infrastructure improvement proposals using a large-scale visual language model (LVLM) and use ontology-driven data filtering to enhance their reliability. This process automatically annotates improvement proposals on pre-accident road images, leading to the construction of a new dataset. Furthermore, we introduce the Baseline approach (OD-RASE model), which leverages LVLM and a diffusion model to produce both infrastructure improvement proposals and generated images of the improved road environment. Our experiments demonstrate that ontology-driven data filtering enables highly accurate prediction of accident-causing road structures and the corresponding improvement plans. We believe that this work contributes to the overall safety of traffic environments and marks an important step toward the broader adoption of autonomous driving systems.
We highlight a curious empirical result within modern Vision Transformers (ViTs). Specifically, self-attention catastrophically fails to train unless it is used in conjunction with a skip connection. This is in contrast to other elements of a ViT that continue to exhibit good performance (albeit suboptimal) when skip connections are removed. Further, we show that this critical dependence on skip connections is a relatively new phenomenon, with previous deep architectures (e.g., CNNs) exhibiting good performance in their absence. In this paper, we theoretically characterize that the self-attention mechanism is fundamentally ill-conditioned and is, therefore, uniquely dependent on skip connections for regularization. Additionally, we propose Token Graying (TG), a simple yet effective complement (to skip connections) that further improves the condition of in- put tokens. We validate our approach in both supervised and self-supervised training methods
We introduce a novel visual tokenization framework that embeds a provable PCA-like structure into the latent token space. While existing visual tokenizers primarily optimize for reconstruction fidelity, they often neglect the structural properties of the latent space--a critical factor for both interpretability and downstream tasks. Our method generates a 1D causal token sequence for images, where each successive token contributes non-overlapping information with mathematically guaranteed decreasing explained variance, analogous to principal component analysis. This structural constraint ensures the tokenizer extracts the most salient visual features first, with each subsequent token adding diminishing yet complementary information. Additionally, we identified and resolved a semantic-spectrum coupling effect that causes the unwanted entanglement of high-level semantic content and low-level spectral details in the tokens by leveraging a diffusion decoder. Experiments demonstrate that our approach achieves state-of-the-art reconstruction performance and enables better interpretability to align with the human vision system. Moreover, autoregressive models trained on our token sequences achieve performance comparable to current state-of-the-art methods while requiring fewer tokens for training and inference.
Prompt-based continual learning provides a rehearsal-free solution by tuning small sets of parameters while keeping pre-trained models frozen. To meet the complex demands of sequential tasks, it is crucial to integrate task-specific knowledge within prompts effectively. However, existing works rely on either fixed learned prompts (i.e., prompts whose representations remain unchanged during new task learning) or on prompts generated from an entangled task-shared space, limiting the representational diversity of the integrated prompt. To address this issue, we propose a novel prompt-evolving mechanism to adaptively aggregate base prompts (i.e., task-specific prompts) into a unified prompt while ensuring diversity. By transforming and aligning base prompts, both previously learned and newly introduced, our approach continuously evolves accumulated knowledge to facilitate learning new tasks. We further introduce a learnable probabilistic gate that adaptively determines which layers to activate during the evolution process. We validate our method on image classification and video action recognition tasks in class-incremental learning, achieving average gains of 9.07% and 7.40% over existing methods across all scenarios.
Diffusion Transformer (DiT) has demonstrated remarkable performance in text-to-image generation; however, its large parameter size results in substantial inference overhead. Existing parameter compression methods primarily focus on pruning, but aggressive pruning often leads to severe performance degradation due to reduced model capacity. To address this limitation, we pioneer the transformation of a dense DiT into a Mixture of Experts (MoE) for structured sparsification, reducing the number of activated parameters while preserving model capacity. Specifically, we replace the Feed-Forward Networks (FFNs) in DiT Blocks with MoE layers, reducing the number of activated parameters in the FFNs by 62.5%.Furthermore, we propose the Mixture of Blocks (MoB) to selectively activate DiT blocks, thereby further enhancing sparsity.To ensure an effective dense-to-MoE conversion, we design a multi-step distillation pipeline, incorporating Taylor metric-based expert initialization, knowledge distillation with load balancing, and group feature loss for MoB optimization. We transform large diffusion transformers (e.g., FLUX.1 [dev]) into an MoE structure, reducing activated parameters by 60% while maintaining original performance and surpassing pruning-based approaches in extensive experiments. Overall, Dense2MoE establishes a new paradigm for efficient text-to-image generation.