CVPR 2025 Papers

Creative remains an inherently abstract concept for both humans and diffusion models. While text-to-image (T2I) diffusion models can easily generate out-of-distribution concepts like "a blue banana", they struggle with generating combinatorial objects such as "a creative mixture that resembles a lettuce and a mantis", due to difficulties in understanding the semantic depth of "creative". Current methods rely heavily on synthesizing reference prompts or images to achieve a creative effect, typically requiring retraining for each unique creative output-a process that is computationally intensive and limits practical applications. To address this, we introduce CreTok, which brings meta-creativity to diffusion models by redefining "creative" as a new token, <CreTok>, thus enhancing models' semantic understanding for combinatorial creativity. CreTok achieves such redefinition by iteratively sampling diverse text pairs from our proposed CangJie dataset to form adaptive prompts and restrictive prompts, and then optimizing the similarity between their respective text embeddings. Extensive experiments demonstrate that <CreTok> enables the universal and direct generation of combinatorial creativity across diverse concepts without additional training, achieving state-of-the-art performance with improved text-image alignment and higher human preference ratings. Code will be made available at https://github.com/fu-feng/CreTok.

Category-agnostic pose estimation aims to locate keypoints on query images according to a few annotated support images for arbitrary novel classes. Existing methods generally extract support features via heatmap pooling, and obtain interacted features from support and query via cross-attention. Hence, these works neglect to mine fine-grained and structure-aware (FGSA) features from both support and query images, which are crucial for pixel-level keypoint localization. To this end, we propose a novel yet concise framework, which recurrently mines FGSA features from both support and query images. Specifically, we design a FGSA mining module based on deformable attention mechanism. On the one hand, we mine fine-grained features by applying deformable attention head over multi-scale feature maps. On the other hand, we mine structure-aware features by offsetting the reference points of keypoints to their linked keypoints. By means of above module, we recurrently mine FGSA features from support and query images, and thus obtain better support features and query estimations. In addition, we propose to use mixup keypoints to pad various classes to a unified keypoint number, which could provide richer supervision than the zero padding used in existing works. We conduct extensive experiments and in-depth studies on large-scale MP-100 dataset, and outperform SOTA method dramatically (+3.2% PCK\@0.05).

Cross-modal retrieval is gaining increasing efficacy and interest from the research community, thanks to large-scale training, novel architectural and learning designs, and its application in LLMs and multimodal LLMs. In this paper, we move a step forward and design an approach that allows for multimodal queries - composed of both an image and a text - and can search within collections of multimodal documents, where images and text are interleaved. Our model, ReT, employs multi-level representations extracted from different layers of both visual and textual backbones, both at the query and document side. To allow for multi-level and cross-modal understanding and feature extraction, ReT employs a novel Transformer-based recurrent cell that integrates both textual and visual features at different layers, and leverages sigmoidal gates inspired by the classical design of LSTMs. Extensive experiments on M2KR and M-BEIR benchmarks show that ReT achieves state-of-the-art performance across diverse settings. Our source code and trained models are publicly available at: https://github.com/aimagelab/ReT.

Classifier-Free Guidance (CFG), which combines the conditional and unconditional score functions with two coefficients summing to one, serves as a practical technique for diffusion model sampling. Theoretically, however, denoising with CFG cannot be expressed as a reciprocal diffusion process, which may consequently leave some hidden risks during use. In this work, we revisit the theory behind CFG and rigorously confirm that the improper configuration of the combination coefficients (*i.e.*, the widely used summing-to-one version) brings about expectation shift of the generative distribution. To rectify this issue, we propose ReCFG with a relaxation on the guidance coefficients such that denoising with ReCFG strictly aligns with the diffusion theory. We further show that our approach enjoys a **closed-form** solution given the guidance strength. That way, the rectified coefficients can be readily pre-computed via traversing the observed data, leaving the sampling speed barely affected. Empirical evidence on real-world data demonstrate the compatibility of our post-hoc design with existing state-of-the-art diffusion models, including both class-conditioned ones (*e.g.*, EDM2 on ImageNet) and text-conditioned ones (*e.g.*, SD3 on CC12M), without any retraining. Code is available at https://github.com/thuxmf/recfg.

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.

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 .

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.

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/.

Existing image editing models struggle to meet realworld demands; despite excelling in academic benchmarks, we are yet to see them adopted to solve real user needs. The datasets that power these models use artificial edits, lacking the scale and ecological validity necessary to address the true diversity of user requests. In response, we introduce REALEDIT, a large-scale image editing dataset with authentic user requests and human-made edits sourced from Reddit. REALEDIT contains a test set of 9.3K examples the community can use to evaluate models on real user requests. Our results show that existing models fall short on these tasks, implying a need for realistic training data. So, we introduce 48K training examples, with which we train our REALEDIT model. Our model achieves substantial gains--outperforming competitors by up to 165 Elo points in human judgment and 92% relative improvement on the automated VIEScore metric on our test set. We deploy our model back on Reddit, testing it on new requests, and receive positive feedback. Beyond image editing, we explore REALEDIT 's potential in detecting edited images by partnering with a deepfake detection non-profit. Finetuning their model on REALEDIT data improves its F1-score by 14 percentage points, underscoring the dataset's value for broad, impactful applications.