CVPR 2025 论文列表

Online stereo rectification is critical for autonomous vehicles and robots in dynamic environments, where factors such as vibration, temperature fluctuations, and mechanical stress can affect rectification accuracy and severely degrade downstream stereo depth estimation. Current dominant approaches for online stereo rectification involve estimating relative camera poses in real time to derive rectification homographies. However, they do not directly optimize for rectification constraints. Additionally, the general-purpose correspondence matchers used in these methods are not trained for rectification, while training of these matchers typically requires ground-truth correspondences which are not available in stereo rectification datasets. To address these limitations, we propose a matching-based stereo rectification framework that is directly optimized for rectification and does not require ground-truth correspondence annotations for training. We assume intrinsics are known as they are generally available on modern devices and are relatively stable. Our framework incorporates a rectification-constrained estimator and applies multi-level, rectification-specific supervision that trains the matcher network for rectification without relying on ground-truth correspondences. Additionally, we create a new rectification dataset with ground-truth optical flow annotations, eliminating bias from evaluation metrics used in prior work that relied on pretrained keypoint matching or optical flow models. Extensive experiments show that our approach outperforms both state-of-the-art matching-based and matching-free methods in vertical flow metric by 10.7% on the Carla-Flowguided dataset and 21.3% on the Semi-Truck Highway dataset, offering superior rectification accuracy.

Understanding 3D motion from videos presents inherent challenges due to the diverse types of movement, ranging from rigid and deformable objects to articulated structures. To overcome this, we propose Liv3Stroke, a novel approach for abstracting objects in motion with deformable 3D strokes. The detailed movements of an object may be represented by unstructured motion vectors or a set of motion primitives using a pre-defined articulation from a template model. Just as a free-hand sketch can intuitively visualize scenes or intentions with a sparse set of lines, we utilize a set of parametric 3D curves to capture a set of spatially smooth motion elements for general objects with unknown structures. We first extract noisy, 3D point cloud motion guidance from video frames using semantic features, and our approach deforms a set of curves to abstract essential motion features as a set of explicit 3D representations. Such abstraction enables an understanding of prominent components of motions while maintaining robustness to environmental factors. Our approach allows direct analysis of 3D object movements from video, tackling the uncertainty that typically occurs when translating real-world motion into recorded footage. The project page is accessible via: https://jaeah.me/liv3stroke_web.

Identifying multiple novel classes in an image, known as open-vocabulary multi-label recognition, is a challenging task in computer vision. Recent studies explore the transfer of powerful vision-language models such as CLIP. However, these approaches face two critical challenges: (1) The local semantics of CLIP are disrupted due to its global pre-training objectives, resulting in unreliable regional predictions. (2) The matching property between image regions and candidate labels has been neglected, relying instead on naive feature aggregation such as average pooling, which leads to spurious predictions from irrelevant regions. In this paper, we present RAM (Recover And Match), a novel framework that effectively addresses the above issues. To tackle the first problem, we propose Ladder Local Adapter (LLA) to enforce the model to refocus on its surrounding local regions, recovering local semantics in a memory-friendly way. For the second issue, we propose Knowledge-Constrained Optimal Transport (KCOT) to suppress meaningless matching to non-GT labels by formulating the task as an optimal transport problem. As a result, RAM achieves state-of-the-art performance on various datasets from three distinct domains, and shows great potential to boost the existing methods. Code: https://github.com/EricTan7/RAM.

Latent diffusion models with Transformer architectures excel at generating high-fidelity images. However, recent studies reveal an optimization dilemma in this two-stage design: while increasing the per-token feature dimension in visual tokenizers improves reconstruction quality, it requires substantially larger diffusion models and more training iterations to achieve comparable generation performance. Consequently, existing systems often settle for sub-optimal solutions, either producing visual artifacts due to information loss within tokenizers or failing to converge fully due to expensive computation costs. We argue that this dilemma stems from the inherent difficulty in learning unconstrained high-dimensional latent spaces. To address this, we propose aligning the latent space with pre-trained vision foundation models when training the visual tokenizers. Our proposed VA-VAE (Vision foundation model Aligned Variational AutoEncoder) significantly expands the reconstruction-generation frontier of latent diffusion models, enabling faster convergence of Diffusion Transformers (DiT) in high-dimensional latent spaces. To exploit the full potential of VA-VAE, we build an enhanced DiT baseline with improved training strategies and architecture designs, termed LightningDiT. The integrated system achieves state-of-the-art (SOTA) performance on ImageNet 256 generation with an FID score of 1.35 while demonstrating remarkable training efficiency by reaching an FID score of 2.11 in just 64 epochs -- representing an over 21 times convergence speedup compared to the original DiT. Models and codes are available at https://github.com/hustvl/LightningDiT.

We present "Humans and Structure from Motion" (HSfM), a method for jointly reconstructing multiple human meshes, scene point clouds, and camera parameters in a metric world coordinate system from a sparse set of uncalibrated multi-view images featuring people. Our approach combines data-driven scene reconstruction with the traditional Structure-from-Motion (SfM) framework to achieve more accurate scene reconstruction and camera estimation while simultaneously recovering human meshes. In contrast to existing scene reconstruction and SfM methods that lack metric scale information, our method estimates approximate metric scale by leveraging the human statistical model. Furthermore, our method reconstructs multiple human meshes within the same world coordinate system with the scene point cloud, effectively capturing spatial relationships among individuals and their positions in the environment. We initialize the reconstruction of humans, scenes, and cameras using robust foundational models and jointly optimize these elements. This joint optimization synergistically improves the accuracy of each component. We compare our method with existing methods on two challenging benchmarks, EgoHumans and EgoExo4D, demonstrating significant improvements in human localization accuracy within the world coordinate frame (reducing error from 3.59m to 1.04m in EgoHumans and from 3.01m to 0.50m in EgoExo4D). Notably, our results show that incorporating human data into the SfM pipeline improves camera pose estimation (e.g., increasing RRA@15 by 20.3% on EgoHumans). Additionally, qualitative results show that our approach improves scene reconstruction quality. Our code is available at muelea.github.io/hsfm.

Reconstructing human-object interactions (HOI) from single images is fundamental in computer vision. Existing methods are primarily trained and tested on indoor scenes due to the lack of 3D data, particularly constrained by the object variety, making it challenging to generalize to real-world scenes with a wide range of objects. The limitations of previous 3D HOI datasets were primarily due to the difficulty in acquiring 3D object assets. However, with the development of 3D reconstruction from single images, recently it has become possible to reconstruct various objects from 2D HOI images.We therefore propose a pipeline for annotating fine-grained 3D humans, objects, and their interactions from single images. We annotated 2.5k+ 3D HOI assets from existing 2D HOI datasets and built the first open-vocabulary in-the-wild 3D HOI dataset Open3DHOI, to serve as a future test set. Moreover, we design a novel Gaussian-HOI optimizer, which efficiently reconstructs the spatial interactions between humans and objects while learning the contact regions.Besides the 3D HOI reconstruction, we also propose several new tasks for 3D HOI understanding to pave the way for future work.

In this paper, we introduce a method for reconstructing 3D humans from a single image using a biomechanically accurate skeleton model. To achieve this, we train a transformer that takes an image as input and estimates the parameters of the model. Due to the lack of training data for this task, we build a pipeline to produce pseudo ground truth model parameters for single images and implement a training procedure that iteratively refines these pseudo labels. Compared to state-of-the-art methods for 3D human mesh recovery, our model achieves competitive performance on standard benchmarks, while it significantly outperforms them in settings with extreme 3D poses and viewpoints. Additionally, we show that previous reconstruction methods frequently violate joint angle limits, leading to unnatural rotations. In contrast, our approach leverages the biomechanically plausible degrees of freedom making more realistic joint rotation estimates. We validate our approach across multiple human pose estimation benchmarks. We make the code, models and data available at: https://isshikihugh.github.io/HSMR/

Due to visual ambiguities and inter-person occlusions, existing human pose estimation methods cannot recover plausible close interactions from in-the-wild videos. Even state-of-the-art large foundation models (e.g., SAM) cannot accurately distinguish human semantics in such challenging scenarios. In this work, we find that human appearance can provide a straightforward cue to address these obstacles. Based on this observation, we propose a dual-branch optimization framework to reconstruct accurate interactive motions with plausible body contacts constrained by human appearances, social proxemics, and physical laws. Specifically, we first train a diffusion model to learn the human proxemic behavior and pose prior knowledge. The trained network and two optimizable tensors are then incorporated into a dual-branch optimization framework to reconstruct human motions and appearances. Several constraints based on 3D Gaussians, 2D keypoints, and mesh penetrations are also designed to assist the optimization. With the proxemics prior and diverse constraints, our method is capable of estimating accurate interactions from in-the-wild videos captured in complex environments. We further build a dataset with pseudo ground-truth interaction annotations, which may promote future research on pose estimation and human behavior understanding. Experimental results on several benchmarks demonstrate that our method outperforms existing approaches. The code and data are available at https://www.buzhenhuang.com/works/CloseApp.html.

The notion of 3D reconstruction as scene understanding is foundational in computer vision. Reconstructing 3D scenes from 2D visual observations necessitates strong priors to disambiguate structure. Much work has been focused on the anthropocentric, which, characterized by smooth surfaces, coherent normals, and regular edges, allows for the integration of strong geometric inductive biases. Here, we consider a more challenging problem where such assumptions do not hold: the reconstruction of natural scenes containing trees, bushes, boulders, and animals. While numerous works have attempted to tackle the problem of reconstructing animals in the wild, they have focused solely on the animal, neglecting environmental context. This limits their usefulness for analysis tasks, as animals exist inherently within the 3D world, and information is lost when environmental factors are disregarded. We propose a method to reconstruct natural scenes from single images. We base our approach on recent advances leveraging the strong world priors ingrained in Large Language Models and train an autoregressive model to decode a CLIP embedding into a structured, compositional scene representation, encompassing both animals and the wild. To enable this, we propose a synthetic dataset comprising one million images and thousands of assets. Through our introduction of a CLIP-projection head, we demonstrate that our approach generalizes to the task of reconstructing animals and their environments in real-world images, despite having been trained solely on synthetic data. We release our dataset and code to encourage future research at https://raw.is.tue.mpg.de/

Plug-and-play (PnP) methods offer an iterative strategy for solving image restoration (IR) problems in a zero-shot manner, using a learned discriminative denoiser as the implicit prior. More recently, a sampling-based variant of this approach, which utilizes a pre-trained generative diffusion model, has gained great popularity for solving IR problems through stochastic sampling. The IR results using PnP with a pre-trained diffusion model demonstrate distinct advantages compared to those using discriminative denoisers, i.e., improved perceptual quality while sacrificing the data fidelity. The unsatisfactory results are due to the lack of integration of these strategies in the IR tasks. In this work, we propose a novel zero-shot IR scheme, dubbed Reconciling Diffusion Model in Dual (RDMD), which leverages only a single pre-trained diffusion model to construct two complementary regularizers. Specifically, the diffusion model in RDMD will iteratively perform deterministic denoising and stochastic sampling, aiming to achieve high-fidelity image restoration with appealing perceptual quality.RDMD also allows users to customize the distortion-perception tradeoff with a single hyperparameter, enhancing the adaptability of the restoration process in different practical scenarios. Extensive experiments on several IR tasks demonstrate that our proposed method could achieve superior results compared to existing approaches on both the FFHQ and ImageNet datasets.

Closed-loop simulation is crucial for end-to-end autonomous driving. Existing sensor simulation methods (e.g., NeRF and 3DGS) reconstruct driving scenes based on conditions that closely mirror training data distributions. However, these methods struggle with rendering novel trajectories, such as lane changes. Recent work, DriveDreamer4D, has demonstrated that integrating world model knowledge alleviates these issues. Although the training-free integration approach is efficient, it still struggles to render larger maneuvers, such as multi-lane shifts.Therefore, we introduce ReconDreamer, which enhances driving scene reconstruction through incremental integration of world model knowledge. Specifically, based on the world model, DriveRestorer is proposed to mitigate ghosting artifacts via online restoration. Additionally, we propose the progressive data update strategy to ensure high-quality rendering for larger maneuvers. Notably, ReconDreamer is the first method to effectively render in large maneuvers (e.g., across multiple lanes, spanning up to 6 meters). Additionally, experimental results demonstrate that ReconDreamer outperforms Street Gaussians in the NTA-IoU, NTL-IoU, and FID, with a relative improvement by 24.87%, 6.72%, and 29.97%. Furthermore, ReconDreamer surpasses DriveDreamer4D with PVG during large maneuver rendering, as verified by a relative improvement of 195.87% in the NTA-IoU metric and a comprehensive user study.

Current Large Multimodal Models (LMMs) empowered visual grounding typically rely on \texttt <SEG> tokens as a text prompt to jointly optimize the vision-language model (e.g., LLaVA) and the downstream task-specific model (e.g., SAM). However, we observe that little research has looked into how it works. In this work, we first visualize the similarity maps, which are obtained by computing the semantic similarity between the \texttt <SEG> token and the image token embeddings derived from the last hidden layer in both the LLaVA encoder and SAM decoder. Intriguingly, we have found that a striking consistency holds in terms of activation responses in the similarity map, which reveals that what the \texttt <SEG> token contributes to is semantic similarity within image-text pairs. Specifically, the \texttt <SEG> token, a placeholder expanded in text vocabulary, extensively queries among individual tokenized image patches to match the semantics of an object from text to the paired image, while the Large Language Models (LLMs) are being fine-tuned. Upon the above findings, we present \toolname, which facilitates LMMs' resilient REAsoning capability of where to attenD under the guidance of highly activated points borrowed from similarity maps. Remarkably, READ features an intuitive design, Similarity as Points module (SasP), which can be seamlessly applied to \texttt <SEG> -like paradigms in a plug-and-play fashion. Also, extensive experiments have been conducted on ReasonSeg and RefCOCO(+/g) datasets. To validate whether READ suffers from catastrophic forgetting of previous skills after fine-tuning, we further assess its generation ability on an augmented FP-RefCOCO(+/g) dataset. All codes and models are publicly available at \href https://github.com/rui-qian/READ https://github.com/rui-qian/READ .

Progress in Embodied AI has made it possible for end-to-end-trained agents to navigate in photo-realistic environments with high-level reasoning and zero-shot or language-conditioned behavior, but evaluations and benchmarks are still dominated by simulation. In this work, we focus on the fine-grained behavior of fast-moving real robots and present a large-scale experimental study involving \numepisodes navigation episodes in a real environment with a physical robot, where we analyze the type of reasoning emerging from end-to-end training. In particular, we study the presence of realistic dynamics which the agent learned for open-loop forecasting, and their interplay with sensing. We analyze the way the agent uses latent memory to hold elements of the scene structure and information gathered during exploration. We probe the planning capabilities of the agent, and find in its memory evidence for somewhat precise plans over a limited horizon. Furthermore, we show in a post-hoc analysis that the value function learned by the agent relates to long-term planning. Put together, our experiments paint a new picture on how using tools from computer vision and sequential decision making have led to new capabilities in robotics and control. An interactive tool is available at https://visual-navigation-reasoning.github.io

This paper pays attention to Weakly Supervised Affordance Grounding (WSAG) task that aims to train model to identify affordance regions using human-object interaction images and egocentric images without the need for costly pixel-level annotations. Most existing methods usually consider the affordance regions to be isolated and directly employ class activation maps to conduct localization, ignoring the relationships with other object components and weakening the performance. For example, a cup's handle is combined with its body to achieve the pouring ability. Obviously, capturing the region relationships is beneficial for improving the localization accuracy of affordance regions. To this end, we first explore exploiting hypergraph to discover these relations and propose a Reasoning Mamba (R-Mamba) framework. We first extract feature embeddings from exo-centric and egocentric images to construct the hypergraphs consisting of multiple vertices and hyperedges, which capture the in-context local region relationships between different visual components. Subsequently, we design a Hypergraph-guided State Space (HSS) block to reorganize these local relationships from the global perspective. By this mechanism, the model could leverage the captured relationships to improve the localization accuracy of affordance regions. Extensive experiments and visualization analyses demonstrate the superiority of our method.

Open-vocabulary 3D visual grounding and reasoning aim to localize objects in a scene based on implicit language descriptions, even when they are occluded. This ability is crucial for tasks such as vision-language navigation and autonomous robotics. However, current methods struggle because they rely heavily on fine-tuning with 3D annotations and mask proposals, which limits their ability to handle diverse semantics and common knowledge required for effective reasoning. To address this, we propose ReasonGrounder, an LVLM-guided framework that uses hierarchical 3D feature Gaussian fields for adaptive grouping based on physical scale, enabling open-vocabulary 3D grounding and reasoning. ReasonGrounder interprets implicit instructions using large vision-language models (LVLM) and localizes occluded objects through 3D Gaussian splatting. By incorporating 2D segmentation masks from the Segment Anything Model (SAM) and multi-view CLIP embeddings, ReasonGrounder selects Gaussian groups based on object scale, enabling accurate localization through both explicit and implicit language understanding, even in novel, occluded views. We also contribute ReasoningGD, a new dataset containing over 10K scenes and 2 million annotations for evaluating open-vocabulary 3D grounding and amodal perception under occlusion. Experiments show that ReasonGrounder significantly improves 3D grounding accuracy in real-world scenarios.

Composed Image Retrieval (CIR) aims to retrieve target images that closely resemble a reference image while integrating user-specified textual modifications, thereby capturing user intent more accurately. Existing training-free zero-shot CIR (ZS-CIR) methods often employ a two-stage process: they first generate a caption for the reference image and then use Large Language Models for reasoning a target description. However, these methods suffer from missing critical visual details and limited reasoning capabilities, leading to suboptimal retrieval performance. To address these challenges, we propose a novel, training-free one-stage method, One-Stage Reflective Chain-of-Thought Reasoning (OSrCIR) for ZS-CIR, which employs Multimodal Large Language Models to retain essential visual information in a single-stage reasoning process, eliminating the information loss in two-stage methods. Our Reflective Chain-of-Thought framework further improves interpretative accuracy by aligning manipulation intent with contextual cues from reference images. OSrCIR achieves performance gains of 1.80% to 6.44% over existing training-free methods across multiple tasks, setting new state-of-the-art results in ZS-CIR and enhancing its utility in vision-language applications. Our code is available at https://github.com/microsoft/ACV/tree/main/OSrCIR.

While computer vision models have made incredible strides in static image recognition, they still do not match human performance in tasks that require the understanding of complex, dynamic motion. This is notably true for real-world scenarios where embodied agents face complex and motion-rich environments. Our approach, BrainNRDS (Brain-Neural Representations of Dynamic Stimuli), leverages state-of-the-art video diffusion models to decouple static image representation from motion generation, enabling us to utilize fMRI brain activity for a deeper understanding of human responses to dynamic visual stimuli. Conversely, we also demonstrate that information about the brain's representation of motion can enhance the prediction of optical flow in artificial systems. Our novel approach leads to four main findings: (1) Visual motion, represented as fine-grained, object-level resolution optical flow, can be decoded from brain activity generated by participants viewing video stimuli; (2) Video encoders outperform image-based models in predicting video-driven brain activity; (3) Brain-decoded motion signals enable realistic video reanimation based only on the initial frame of the video; and (4) We extend prior work to achieve full video decoding from video-driven brain activity. BrainNRDS advances our understanding of how the brain represents spatial and temporal information in dynamic visual scenes. Our findings demonstrate the potential of combining brain imaging with video diffusion models for developing more robust and biologically-inspired computer vision systems. We show additional decoding and encoding examples on this site: https://brain-nrds.github.io/.

The zero-shot capabilities of Vision-Language Models (VLMs) have been widely leveraged to improve predictive performance. However, previous works on transductive or test-time adaptation (TTA) often make strong assumptions about the data distribution, such as the presence of all classes. Our work challenges these favorable deployment scenarios and introduces a more realistic evaluation framework, including (i) a variable number of effective classes for adaptation within a single batch, and (ii) non-i.i.d. batches of test samples in online adaptation settings. We provide comprehensive evaluations, comparisons, and ablation studies that demonstrate how current transductive or TTA methods for VLMs systematically compromise the models' initial zero-shot robustness across various realistic scenarios, favoring performance gains under advantageous assumptions about the test sample distributions. Furthermore, we introduce StatA, a versatile method that can handle a wide range of deployment scenarios, including those with a variable number of effective classes at test time. Our approach incorporates a novel regularization term designed specifically for VLMs, which acts as a statistical anchor preserving the initial text-encoder knowledge, particularly in low-data regimes. Code available at https://github.com/MaxZanella/StatA.