CVPR 2025 Papers

Agglomerative models have recently emerged as a powerful approach to training vision foundation models, leveraging multi-teacher distillation from existing models such as CLIP, DINO, and SAM. This strategy enables the efficient creation of robust models, combining the strengths of individual teachers while significantly reducing computational and resource demands. In this paper, we thoroughly analyze state-of-the-art agglomerative models, identifying critical challenges including resolution mode shifts, teacher imbalance, idiosyncratic teacher artifacts, and an excessive number of output tokens. To address these issues, we propose several novel solutions: multi-resolution training, mosaic augmentation, and improved balancing of teacher loss functions. Specifically, in the context of Vision Language Models, we introduce a token compression technique to maintain high-resolution information within a fixed token count. We release our top-performing variants at multiple scales (-B, -L, -H, and -g), along with inference code and pretrained weights.

Diffusion models have achieved remarkable success in image generation, with applications broadening across various domains.Inpainting is one such application that can benefit significantly from diffusion models. Existing methods either hijack the reverse process of a pretrained diffusion model or cast the problem into a larger framework, i.e., conditioned generation. However, these approaches often require nested loops in the generation process or additional components for conditioning. In this paper, we present region-aware diffusion models (RAD) for inpainting with a simple yet effective reformulation of the vanilla diffusion models. RAD utilizes a different noise schedule for each pixel, which allows local regions to be generated asynchronously while considering the global image context. A plain reverse process requires no additional components, enabling RAD to achieve inference time up to 100 times faster than the state-of-the-art approaches. Moreover, we employ low-rank adaptation (LoRA) to fine-tune RAD based on other pretrained diffusion models, reducing computational burdens in training as well. Experiments demonstrated that RAD provides state-of-the-art results both qualitatively and quantitatively, on the FFHQ, LSUN Bedroom, and ImageNet datasets.

This paper explores using large language models (LLMs) as low-level action planners for embodied tasks. While LLMs excel as the robot's "brain" for high-level planning, they face challenges in directly controlling the "body" by generating precise low-level actions. This limitation arises from LLMs' strength in high-level conceptual understanding but their inability to handle spatial perception effectively, restricting their potential in embodied tasks. To address this, we bridge the gap by enabling LLMs to not only comprehend complex instructions but also produce actionable, low-level plans. We introduce Room to Chessboard (R2C), a novel semantic representation that maps environmental states onto a grid-based chessboard, empowering LLMs to generate specific low-level coordinates and guide the robot in a manner akin to playing a game of chess. To further enhance decision-making, we propose the Chain-of-Thought Decision (CoT-D) paradigm, which improves LLMs' interpretability and context-awareness in spatial reasoning. By jointly training LLMs for high-level task decomposition and low-level action generation, we create a unified "brain-body" system capable of handling complex, free-form instructions while producing precise low-level actions, allowing the robot to flexibly control its movements and adapt to varying tasks. We validate R2C using both fine-tuned open-source LLMs and GPT-4, demonstrating effectiveness on the challenging ALFRED benchmark. Results show that with our R2C framework, LLMs can effectively act as low-level planners, generalizing across diverse settings and open-vocabulary robotic tasks. The code and demonstrations are available at: https://vipl-vsu.github.io/Room2Chessboard.

Vision-language models (VLMs), such as CLIP, have gained significant popularity as foundation models, with numerous fine-tuning methods developed to enhance performance on downstream tasks. However, due to their inherent vulnerability and the common practice of selecting from a limited set of open-source models, VLMs suffer from a higher risk of adversarial attacks than traditional vision models. Existing defense techniques typically rely on adversarial fine-tuning during training, which requires labeled data and lacks of flexibility for downstream tasks. To address these limitations, we propose robust test-time prompt tuning (R-TPT), which mitigates the impact of adversarial attacks during the inference stage. We first reformulate the classic marginal entropy objective by eliminating the term that introduces conflicts under adversarial conditions, retaining only the pointwise entropy minimization. Furthermore, we introduce a plug-and-play reliability-based weighted ensembling strategy, which aggregates useful information from reliable augmented views to strengthen the defense. R-TPT enhances defense against adversarial attacks without requiring labeled training data while offering high flexibility for inference tasks. Extensive experiments on widely used benchmarks with various attacks demonstrate the effectiveness of R-TPT.

Learning-based visual localization methods that use scene coordinate regression (SCR) offer the advantage of smaller map sizes. However, on datasets with complex illumination changes or image-level ambiguities, it remains a less robust alternative to feature matching methods. This work aims to close the gap. We introduce a covisibility graph-based global encoding learning and data augmentation strategy, along with a depth-adjusted reprojection loss to facilitate implicit triangulation. Additionally, we revisit the network architecture and local feature extraction module. Our method achieves state-of-the-art on challenging large-scale datasets without relying on network ensembles or 3D supervision. On Aachen Day-Night, we are 10x more accurate than previous SCR methods with similar map sizes and require at least 5x smaller map sizes than any other SCR method while still delivering superior accuracy. Code is available at: https://github.com/cvg/scrstudio.

Visual Prompt Learning (VPL) has emerged as a powerful strategy for harnessing the capabilities of large-scale pre-trained models (PTMs) to tackle specific downstream tasks. However, the opaque nature of PTMs in many real-world applications has led to a growing interest in gradient-free approaches within VPL. A significant challenge with existing black-box VPL methods lies in the high dimensionality of visual prompts, which necessitates considerable API queries for tuning, thereby impacting efficiency. To address this issue, we propose a novel query-efficient framework for black-box visual prompting, designed to generate input-dependent visual prompts efficiently for large-scale black-box PTMs. Our framework is built upon the insight of reparameterizing prompts using neural networks, improving the typical pre-training-fine-tuning paradigm through the subspace learning strategy to maximize efficiency and adaptability from both the perspective of initial weights and parameter dimensionality. This tuning intrinsically optimizes low-dimensional representations within the well-learned subspace, enabling the efficient adaptation of the network to downstream tasks. Our approach significantly reduces the necessity for substantial API queries to PTMs, presenting an efficient method for leveraging large-scale black-box PTMs in visual prompting tasks. Most experimental results across various benchmarks demonstrate the effectiveness of our method, showcasing substantial reductions in the number of required API queries to PTMs while maintaining or even enhancing performance on downstream tasks.

Monocular Depth Estimation (MDE) has emerged as a pivotal task in computer vision, supporting numerous real-world applications. However, deploying accurate depth estimation models on resource-limited edge devices, especially Application-Specific Integrated Circuits (ASICs), is challenging due to the high computational and memory demands. Recent advancements in foundational depth estimation deliver impressive results but further amplify the difficulty of deployment on ASICs. To address this, we propose QuartDepth which adopts post-training quantization to quantize MDE models with hardware accelerations for ASICs. Our approach involves quantizing both weights and activations to 4-bit precision, reducing the model size and computation cost. To mitigate the performance degradation, we introduce activation polishing and compensation algorithm applied before and after activation quantization, as well as a weight reconstruction method for minimizing errors in weight quantization. Furthermore, we design a flexible and programmable hardware accelerator by supporting kernel fusion and customized instruction programmability, enhancing throughput and efficiency. Experimental results demonstrate that our framework achieves competitive accuracy while enabling fast inference and higher energy efficiency on ASICs, bridging the gap between high-performance depth estimation and practical edge-device applicability. Code: https://github.com/shawnricecake/quart-depth

Deep neural networks, while achieving remarkable success across diverse tasks, demand significant resources, including computation, GPU memory, bandwidth, storage, and energy. Network quantization, as a standard compression and acceleration technique, reduces storage costs and enables potential inference acceleration by discretizing network weights and activations into a finite set of integer values. However, current quantization methods are often complex and sensitive, requiring extensive task-specific hyperparameters, where even a single misconfiguration can impair model performance, limiting generality across different models and tasks. In this paper, we propose Quantization without Tears (QwT), a method that simultaneously achieves quantization speed, accuracy, simplicity, and generality. The key insight of QwT is to incorporate a lightweight additional structure into the quantized network to mitigate information loss during quantization. This structure consists solely of a small set of linear layers, keeping the method simple and efficient. More importantly, it provides a closed-form solution, allowing us to improve accuracy effortlessly under 2 minutes. Extensive experiments across various vision, language, and multimodal tasks demonstrate that QwT is both highly effective and versatile. In fact, our approach offers a robust solution for network quantization that combines simplicity, accuracy, and adaptability, which provides new insights for the design of novel quantization paradigms.

Realistic hair motion is crucial for high-quality avatars, but it is often limited by the computational resources available for real-time applications. To address this challenge, we propose a novel neural approach to predict physically plausible hair deformations that generalizes to various body poses, shapes, and hair styles. Our model is trained using a self-supervised loss, eliminating the need for expensive data generation and storage. We demonstrate our method's effectiveness through numerous results across a wide range of pose and shape variations, showcasing its robust generalization capabilities and temporally smooth results. Our approach is highly suitable for real-time applications with an inference time of only a few milliseconds on consumer hardware and its ability to scale to predicting 1000 grooms in 0.3 seconds. Code will be released.

Defocus deblurring is a challenging task due to the spatially varying blur. Recent works have shown impressive results in data-driven approaches using dual-pixel (DP) sensors. Quad-pixel (QP) sensors represent an advanced evolution of DP sensors, providing four distinct sub-aperture views in contrast to only two views offered by DP sensors. However, research on QP-based defocus deblurring is scarce. In this paper, we propose a novel end-to-end learning-based approach for defocus deblurring that leverages QP data. To achieve this, we design a QP defocus and all-in-focus image pair acquisition method and provide a QP Defocus Deblurring (QPDD) dataset containing 4,935 image pairs. We then introduce a Local-gate assisted Mamba Network (LMNet), which includes a two-branch encoder and a Simple Fusion Module (SFM) to fully utilize features of sub-aperture views. In particular, our LMNet incorporates a Local-gate assisted Mamba Block (LAMB) that mitigates local pixel forgetting and channel redundancy within Mamba, and effectively captures global and local dependencies. By extending the defocus deblurring task from a DP-based to a QP-based approach, we demonstrate significant improvements in restoring sharp images. Comprehensive experimental evaluations further indicate that our approach outperforms state-of-the-art methods.

Burst super-resolution (BurstSR) aims to reconstruct high-resolution images by fusing subpixel details from multiple low-resolution burst frames. The primary challenge lies in effectively extracting useful information while mitigating the impact of high-frequency noise. Most existing methods rely on frame-by-frame fusion, which often struggles to distinguish informative subpixels from noise, leading to suboptimal performance. To address these limitations, we introduce a novel Query Mamba Burst Super-Resolution (QMambaBSR) network. Specifically, we observe that sub-pixels have consistent spatial distribution while noise appears randomly. Considering the entire burst sequence during fusion allows for more reliable extraction of consistent subpixels and better suppression of noise outliers. Based on this, a Query State Space Model (QSSM) is proposed for both inter-frame querying and intra-frame scanning, enabling a more efficient fusion of useful subpixels. Additionally, to overcome the limitations of static upsampling methods that often result in over-smoothing, we propose an Adaptive Upsampling (AdaUp) module that dynamically adjusts the upsampling kernel to suit the characteristics of different burst scenes, achieving superior detail reconstruction. Extensive experiments on four benchmark datasets--spanning both synthetic and real-world images--demonstrate that QMambaBSR outperforms existing state-of-the-art methods.

In this work, we address the challenge of adaptive pediatric Left Ventricular Ejection Fraction (LVEF) assessment. While Test-time Training (TTT) approaches show promise for this task, they suffer from two significant limitations. Existing TTT works are primarily designed for classification tasks rather than continuous value regression, and they lack mechanisms to handle the quasi-periodic nature of cardiac signals. To tackle these issues, we propose a novel Quasi-Periodic Adaptive Regression with Test-time Training (Q-PART) framework. In the training stage, the proposed Quasi-Period Network decomposes the echocardiogram into periodic and aperiodic components within latent space by combining parameterized helix trajectories with Neural Controlled Differential Equations. During inference, our framework further employs a variance minimization strategy across image augmentations that simulate common quality issues in echocardiogram acquisition, along with differential adaptation rates for periodic and aperiodic components. Theoretical analysis is provided to demonstrate that our variance minimization objective effectively bounds the regression error under mild conditions. Furthermore, extensive experiments across three pediatric age groups demonstrate that Q-PART not only significantly outperforms existing approaches in pediatric LVEF prediction, but also exhibits strong clinical screening capability with high mAUROC scores (up to 0.9747) and maintains gender-fair performance across all metrics, validating its robustness and practical utility in pediatric echocardiography analysis.

Evaluating text-to-vision content hinges on two crucial aspects: **visual quality** and **alignment**. While significant progress has been made in developing objective models to assess these dimensions, the performance of such models heavily relies on the scale and quality of human annotations. According to **Scaling Law**, increasing the number of human-labeled instances follows a predictable pattern that enhances the performance of evaluation models.Therefore, we introduce a comprehensive dataset designed to **E**valuate **V**isual quality and **A**lignment **L**evel for text-to-vision content (**Q-EVAL-100K**), featuring the largest collection of human-labeled Mean Opinion Scores (MOS) for the mentioned two aspects.The **Q-EVAL-100K** dataset encompasses both text-to-image and text-to-video models, with 960K human annotations specifically focused on visual quality and alignment for 100K instances (60K images and 40K videos). Leveraging this dataset with context prompt, we propose **Q-Eval-Score**, a unified model capable of evaluating both visual quality and alignment with special improvements for handling long-text prompt alignment.Experimental results indicate that the proposed **Q-Eval-Score** achieves superior performance on both visual quality and alignment, with strong generalization capabilities across other benchmarks. These findings highlight the significant value of the **Q-EVAL-100K** dataset. **The data and code will be released** to help promote the generation models.

Recent advancements in diffusion models, particularly the architectural transformation from UNet-based models to Diffusion Transformers (DiTs), significantly improve the quality and scalability of image and video generation. However, despite their impressive capabilities, the substantial computational costs of these large-scale models pose significant challenges for real-world deployment. Post-Training Quantization (PTQ) emerges as a promising solution, enabling model compression and accelerated inference for pretrained models, without the costly retraining. However, research on DiT quantization remains sparse, and existing PTQ frameworks, primarily designed for traditional diffusion models, tend to suffer from biased quantization, leading to notable performance degradation. In this work, we identify that DiTs typically exhibit significant spatial variance in both weights and activations, along with temporal variance in activations. To address these issues, we propose Q-DiT, a novel approach that seamlessly integrates two key techniques: automatic quantization granularity allocation to handle the significant variance of weights and activations across input channels, and sample-wise dynamic activation quantization to adaptively capture activation changes across both timesteps and samples. Extensive experiments conducted on ImageNet and VBench demonstrate the effectiveness of the proposed Q-DiT. Specifically, when quantizing DiT-XL/2 to W6A8 on ImageNet (256 x256), Q-DiT achieves a remarkable reduction in FID by 1.09 compared to the baseline. Under the more challenging W4A8 setting, it maintains high fidelity in image and video generation, establishing a new benchmark for efficient, high-quality quantization in DiTs.

With the rising interest in research on Large Multi-modal Models (LMMs) for video understanding, many studies have emphasized general video comprehension capabilities, neglecting the systematic exploration into video quality understanding. To address this oversight, we introduce Q-Bench-Video in this paper, a new benchmark specifically designed to evaluate LMMs' proficiency in discerning video quality. a) To ensure video source diversity, Q-Bench-Video encompasses videos from natural scenes, AI-generated Content (AIGC), and Computer Graphics (CG). b) Building on the traditional multiple-choice questions format with the Yes-or-No and What-How categories, we include Open-ended questions to better evaluate complex scenarios. Additionally, we incorporate the video pair quality comparison question to enhance comprehensiveness. c) Beyond the traditional Technical, Aesthetic, and Temporal distortions, we have expanded our evaluation aspects to include the dimension of AIGC distortions, which addresses the increasing demand for video generation. Finally, we collect a total of 2,378 question-answer pairs and test them on 12 open-source & 5 proprietary LMMs. Our findings indicate that while LMMs have a foundational understanding of video quality, their performance remains incomplete and imprecise, with a notable discrepancy compared to human performance. Through Q-Bench-Video, we seek to catalyze community interest, stimulate further research, and unlock the untapped potential of LMMs to close the gap in video quality understanding.

We propose PyTorchGeoNodes, a differentiable module for reconstructing 3D objects and their parameters from images using interpretable shape programs. Unlike traditional CAD model retrieval, shape programs allow reasoning about semantic parameters, editing, and a low memory footprint. Despite their potential, shape programs for 3D scene understanding have been largely overlooked. Our key contribution is enabling gradient-based optimization by parsing shape programs, or more precisely procedural models designed in Blender, into efficient PyTorch code. While there are many possible applications of our PyTochGeoNodes, we show that a combination of PyTorchGeoNodes with genetic algorithm is a method of choice to optimize both discrete and continuous shape program parameters for 3D reconstruction and understanding of 3D object parameters. Our modular framework can be further integrated with other reconstruction algorithms, and we demonstrate one such integration to enable procedural Gaussian splatting. Our experiments on the ScanNet dataset show that our method achieves accurate reconstructions while enabling, until now, unseen level of 3D scene understanding.

Video virtual try-on aims to seamlessly dress a subject in a video with a specific garment. The primary challenge involves preserving the visual authenticity of the garment while dynamically adapting to the pose and physique of the subject. While existing methods have predominantly focused on image-based virtual try-on, extending these techniques directly to videos often results in temporal inconsistencies. Most current video virtual try-on approaches alleviate this challenge by incorporating temporal modules, yet still overlook the critical spatiotemporal pose interactions between human and garment. Effective pose interactions in videos should not only consider spatial alignment between human and garment poses in each frame but also account for the temporal dynamics of human poses throughout the entire video. With such motivation, we propose a new framework, namely Dynamic Pose Interaction Diffusion Models (DPIDM), to leverage diffusion models to delve into dynamic pose interactions for video virtual try-on. Technically, DPIDM introduces a skeleton-based pose adapter to integrate synchronized human and garment poses into the denoising network. A hierarchical attention module is then exquisitely designed to model intra-frame human-garment pose interactions and long-term human pose dynamics across frames through pose-aware spatial and temporal attention mechanisms. Moreover, DPIDM capitalizes on a temporal regularized attention loss between consecutive frames to enhance temporal consistency. Extensive experiments conducted on VITON-HD, VVT and ViViD datasets demonstrate the superiority of our DPIDM against the baseline methods. Notably, DPIDM achieves VFID score of 0.506 on VVT dataset, leading to 60.5% improvement over the state-of-the-art GPD-VVTO approach.

Thermal object detection is a critical task in various fields, such as surveillance and autonomous driving. Current state-of-the-art (SOTA) models always leverage a prior Thermal-To-Visible (T2V) translation model to obtain visible spectrum information, followed by a cross-modality aggregation module to fuse information from both modalities. However, this fusion approach does not fully exploit the complementary visible spectrum information beneficial for thermal detection. To address this issue, we propose a novel cross-modal fusion method called Pseudo Visible Feature Fine-Grained Fusion (PFGF). Specifically, a graph is constructed with nodes generated from multi-level thermal features and pseudo-visual latent features produced by the T2V model. Each level of features corresponds to a subgraph. An Inter-Mamba block is proposed to perform cross-modality fusion between nodes at the lowest level; while a Cascade Knowledge Integration (CKI) strategy is used to fuse low-level fused information to high-level subgraphs in a cascade manner. After several iterations of graph node updating, each subgraph outputs an aggregated feature to the detection head respectively. Unlike previous cross-modal fusion methods, our approach explicitly models high-level relationships between cross-modal data, effectively fusing different granularity information. Experimental results demonstrate that our method achieves SOTA detection performance. Code is available at https://github.com/liting1018/PFGF.

Embodied intelligence requires agents to interact with 3D environments in real time based on language instructions. A foundational task in this domain is ego-centric 3D visual grounding. However, the point clouds rendered from RGB-D images retain a large amount of redundant background data and inherent noise, both of which can interfere with the manifold structure of the target regions. Existing point cloud enhancement methods often require a tedious process to improve the manifold, which is not suitable for real-time tasks. We propose Proxy Transformation suitable for multimodal task to efficiently improve the point cloud manifold. Our method first leverages Deformable Point Clustering to identify the point cloud sub-manifolds in target regions. Then, we propose a Proxy Attention module that utilizes multimodal proxies to guide point cloud transformation. Built upon Proxy Attention, we design a submanifold transformation generation module where textual information globally guides translation vectors for different submanifolds, optimizing relative spatial relationships of target regions. Simultaneously, image information guides linear transformations within each submanifold, refining the local point cloud manifold of target regions. Extensive experiments demonstrate that Proxy Transformation significantly outperforms all existing methods, achieving an impressive improvement of 7.49% on easy targets and 4.60% on hard targets, while reducing the computational overhead of attention blocks by 40.6%. These results establish a new SOTA in ego-centric 3D visual grounding, showcasing the effectiveness and robustness of our approach.