CVPR 2025 Papers

We introduce XTRA, a vision model pre-trained with a novel auto-regressive objective that significantly enhances both sample and parameter efficiency compared to previous auto-regressive image models. Unlike contrastive or masked image modeling methods, which have not been demonstrated as having consistent scaling behavior on unbalanced internet data,auto-regressive vision models exhibit scalable and promising performance as model and dataset size increase. In contrast to standard auto-regressive models, XTRA employs a Block Causal Mask, where each Block represents k x k tokens rather than relying on a standard causal mask. By reconstructing pixel values block by block, XTRA captures higher-level structural patterns over larger image regions. Predicting on blocks allows the model to learn relationships across broader areas of pixels, enabling more abstract and semantically meaningful representations than traditional next-token prediction.This simple modification yields two key results. First, XTRA is sample-efficient. Despite being trained on 152x fewer samples (13.1M vs. 2B), XTRA ViT-H/14 surpasses the top-1 average accuracy of the previous state-of-the-art auto-regressive model across 15 diverse image recognition benchmarks. Second, XTRA is parameter-efficient. Compared to auto-regressive models trained on ImageNet-1k, XTRA ViT-B/16 outperforms in linear and attentive probing tasks, using 7-16x fewer parameters (85M vs. 1.36B/0.63B).

Global effective receptive field plays a crucial role for image style transfer (ST) to obtain high-quality stylized results. However, existing ST backbones (e.g., CNNs and Transformers) suffer huge computational complexity to achieve global receptive fields. Recently, the State Space Model (SSM), especially the improved variant Mamba, has shown great potential for long-range dependency modeling with linear complexity, which offers a approach to resolve the above dilemma. In this paper, we develop a Mamba-based style transfer framework, termed SaMam. Specifically, a mamba encoder is designed to efficiently extract content and style information. In addition, a style-aware mamba decoder is developed to flexibly adapt to various styles. Moreover, to address the problems of local pixel forgetting, channel redundancy and spatial discontinuity of existing SSMs, we introduce both local enhancement and zigzag scan. Qualitative and quantitative results demonstrate that our SaMam outperforms state-of-the-art methods in terms of both accuracy and efficiency.

The rapid proliferation of large-scale text-to-image diffusion (T2ID) models has raised serious concerns about their potential misuse in generating harmful content. Although numerous methods have been proposed for erasing undesired concepts from T2ID models, they often provide a false sense of security; concept-erased models (CEMs) can still be manipulated via adversarial attacks to regenerate the erased concept. While a few robust concept erasure methods based on adversarial training have emerged recently, they compromise on utility (generation quality for benign concepts) to achieve robustness and/or remain vulnerable to advanced embedding space attacks. These limitations stem from the failure of robust CEMs to thoroughly search for blind spots in the embedding space. To bridge this gap, we propose STEREO, a novel two-stage framework that employs adversarial training as a first step rather than the only step for robust concept erasure. In the first stage, STEREO employs adversarial training as a vulnerability identification mechanism to search thoroughly enough. In the second robustly erase once stage, STEREO introduces an anchor-concept-based compositional objective to robustly erase the target concept in a single fine-tuning stage, while minimizing the degradation of model utility. We benchmark STEREO against seven state-of-the-art concept erasure methods, demonstrating its superior robustness to both white-box and black-box attacks, while largely preserving utility.

Recent text-to-image (T2I) diffusion models have demonstrated remarkable capabilities in visual synthesis, yet their performance heavily relies on the quality of input prompts. However, optimizing discrete prompts remains challenging because the discrete nature of tokens prevents the direct application of the gradient descent method and the vast search space of possible token combinations. As a result, existing approaches either suffer from quantization errors when employing continuous optimization techniques or become trapped in local optima due to coordinate-wise greedy search. In this paper, we propose STEPS, a novel Sequential probability Tensor Estimation approach for hard Prompt Search. Our method reformulates discrete prompt optimization as a sequential probability tensor estimation problem, leveraging the inherent low-rank characteristics to address the curse of dimensionality. To further improve the computational efficiency, we develop a memory-bounded sampling approach that shrinks the prompt space without the iteration step dependency while preserving sequential optimization dynamics. Extensive experiments on various public datasets demonstrate that our method consistently outperforms existing approaches in T2I generation, cross-model prompt transferability, and harmful prompt optimization, validating the effectiveness of the proposed framework.

Diffusion probabilistic model is becoming the cornerstone of data generation, especially generating high-quality images. As an extension, video diffusion generation is in urgent need of a principled temporal-sequence diffusion way, while the spatial-domain diffusion dominates most video diffusion methods. In this work, we propose an explicit Spatio-Temporal Dual Diffusion (STDD) method by principledly extending the standard diffusion model to a spatio-temporal diffusion model for joint spatial and temporal noise propagation/reduction. Mathematically, an analysable dual diffusion process is derived to accumulate noises/information in temporal sequence as well as spatial domain. Correspondingly, we theoretically derive a spatio-temporal probabilistic reverse diffusion process and propose an accelerated sampling way to reduce the inference cost. In principle, the spatio-temporal dual diffusion enables the information of previous frames to be transferred to the current frame, which thus could be beneficial for video consistency. Extensive experiments demonstrate that our proposed STDD is more competitive over the state-of-the-art methods in the task of video generation/prediction as well as text-to-video generation.

Generating layouts from textual descriptions by large language models (LLMs) plays a crucial role in precise spatial reasoning-induced domains such as robotic object rearrangement and text-to-image generation. However, current methods face challenges in limited real-world examples, handling diverse layout descriptions and varying levels of granularity. To address these issues, a novel framework named Spatial Knowledge Enhanced Layout (SKE-Layout), is introduced. SKE-Layout integrates mixed spatial knowledge sources, leveraging both real and synthetic data to enhance spatial contexts. It utilizes diverse representations tailored to specific tasks and employs contrastive learning and multitask learning techniques for accurate spatial knowledge retrieval. This framework generates more accurate and fine-grained visual layouts for object rearrangement and text-to-image generation tasks, achieving improvements of 5%-30% compared to existing methods.

Controllable generation has achieved substantial progress in both 2D and 3D domains, yet current conditional generation methods still face limitations in describing detailed shape structures. Skeletons can effectively represent and describe object anatomy and pose. Unfortunately, past studies are often limited to human skeletons. In this work, we generalize skeletal conditioned generation to arbitrary structures. First, we design a reliable mesh skeletonization pipeline to generate a large-scale mesh-skeleton paired dataset.Based on the dataset, a multi-view and 3D generation pipeline is built. We propose to represent 3D skeletons by Coordinate Color Encoding as 2D conditional images. A Skeletal Correlation Module is designed to extract global skeletal features for condition injection. After multi-view images are generated, 3D assets can be obtained by incorporating a large reconstruction model, followed by a UV texture refinement stage. As a result, our method achieves instant generation of multi-view and 3D contents that are aligned with given skeletons. The proposed techniques largely improve the object-skeleton alignment and generation quality.

Large Multimodal Models (LMMs) have demonstrated impressive capabilities in multimodal understanding and generation, pushing forward advancements in text-to-image generation.However, achieving accurate text-image alignment for LMMs, particularly in compositional scenarios, remains challenging. Existing approaches, such as layout planning for multi-step generation and learning from human feedback or AI feedback, depend heavily on prompt engineering, costly human annotations, and continual upgrading, limiting flexibility and scalability. In this work, we introduce a model-agnostic iterative self-improvement framework (**SILMM**) that can enable LMMs to provide helpful and scalable self-feedback and optimize text-image alignment via Direct Preference Optimization (DPO). DPO can readily applied to LMMs that use discrete visual tokens as intermediate image representations; while it is less suitable for LMMs with continuous visual features, as obtaining generation probabilities is challenging.To adapt SILMM to LMMs with continuous features, we propose a diversity mechanism to obtain diverse representations and a kernel-based continuous DPO for alignment. Extensive experiments on three compositional text-to-image generation benchmarks validate the effectiveness and superiority of SILMM, showing improvements exceeding 30% on T2I-CompBench++ and around 20% on DPG-Bench.

Interactive segmentation is to segment the mask of the target object according to the user's interactive prompts. There are two mainstream strategies: early fusion and late fusion. Current specialist models utilize the early fusion strategy that encodes the combination of images and prompts to target the prompted objects, yet repetitive complex computations on the images result in high latency. Late fusion models extract image embeddings once and merge them with the prompts in later interactions. This strategy avoids redundant image feature extraction and improves efficiency significantly. A recent milestone is the Segment Anything Model (SAM). However, this strategy limits the models ability to extract detailed information from the prompted target zone. To address this issue, we propose SAM-REF, a two-stage refinement framework that fully integrates images and prompts by using a lightweight refiner into the interaction of late fusion, which combines the accuracy of early fusion and maintains the efficiency of late fusion. Through extensive experiments, we show that our SAM-REF model outperforms the current state-of-the-art method in most metrics on segmentation quality without compromising efficiency.

Text-driven motion generation has advanced significantly with the rise of denoising diffusion models. However, previous methods often oversimplify representations for the skeletal joints, temporal frames, and textual words, limiting their ability to fully capture the information within each modality and their interactions. Moreover, when using pre-trained models for downstream tasks, such as editing, they typically require additional efforts, including manual interventions, optimization, or fine-tuning. In this paper, we introduce a skeleton-aware latent diffusion (SALAD), a model that explicitly captures the intricate inter-relationships between joints, frames, and words. Furthermore, by leveraging cross-attention maps produced during the generation process, we enable the attention-based zero-shot text-driven motion editing using a pre-trained SALAD model, requiring no additional user input beyond text prompts. Our approach significantly outperforms previous methods in terms of text-motion alignment without compromising generation quality, and demonstrates practical versatility by providing diverse editing capabilities beyond generation. Code is available at project page.

Infrared Small Target Detection (IRSTD) aims to identify low signal-to-noise ratio small targets in infrared images with complex backgrounds, which is crucial for various applications. However, existing IRSTD methods typically rely solely on image modalities for processing, which fail to fully capture contextual information, leading to limited detection accuracy and adaptability in complex environments. Inspired by vision-language models, this paper proposes a novel framework, SAIST, which integrates textual information with image modalities to enhance IRSTD performance. The framework consists of two main components: Scene Recognition Contrastive Language-Image Pretraining (SR-CLIP) and CLIP-guided Segment Anything Model (CG-SAM). SR-CLIP generates a set of visual descriptions through object-object similarity and object-scene relevance, embedding them into learnable prompts to refine the textual description set. This reduces the domain gap between vision and language, generating precise textual and visual prompts. CG-SAM utilizes the prompts generated by SR-CLIP to accurately guide the Mask Decoder in learning prior knowledge of background features, while incorporating infrared imaging equations to improve small target recognition in complex backgrounds and significantly reduce the false alarm rate. Additionally, this paper introduces the first multimodal IRSTD dataset, MIRSTD, which contains abundant image-text pairs. Experimental results demonstrate that the proposed SAIST method outperforms existing state-of-the-art approaches.

Deep learning-based image registration methods have shown state-of-the-art performance and rapid inference speeds. Despite these advances, many existing approaches fall short in capturing spatially varying information in non-local regions of feature maps due to the reliance on spatially-shared convolution kernels. This limitation leads to suboptimal estimation of deformation fields. In this paper, we propose a 3D Spatial-Awareness Convolution Block (SACB) to enhance the spatial information within feature representations. Our SACB estimates the spatial clusters within feature maps by leveraging feature similarity and subsequently parameterizes the adaptive convolution kernels across diverse regions. This adaptive mechanism generates the convolution kernels (weights and biases) tailored to spatial variations, thereby enabling the network to effectively capture spatially varying information. Building on SACB, we introduce a pyramid flow estimator (named SACB-Net) that integrates SACBs to facilitate multi-scale flow composition, particularly addressing large deformations. Experimental results on the brain IXI and LPBA datasets as well as Abdomen CT datasets demonstrate the effectiveness of SACB and the superiority of SACB-Net over the state-of-the-art learning-based registration methods. The code is available at https://github.com/x-xc/SACB_Net.

In this paper, we present S2D-LFE, an innovative approach for sparse-to-dense light field event generation. For the first time to our knowledge, S2D-LFE enables controllable novel view synthesis only from sparse-view light field event (LFE) data, and addresses three critical challenges for the LFE generation task: simplicity, controllability, and consistency. The simplicity aspect eliminates the dependency on frame-based modality, which often suffers from motion blur and low frame-rate limitations. The controllability aspect enables precise view synthesis under sparse LFE conditions with view-related constraints. The consistency aspect ensures both cross-view and temporal coherence in the generated results. To realize S2D-LFE, we develop a novel diffusion-based generation network with two key components. First, we design an LFE-customized variational auto-encoder that effectively compresses and reconstructs LFE by integrating cross-view information. Second, we design an LFE-aware injection adaptor to extract comprehensive geometric and texture priors. Furthermore, we construct a large-scale synthetic LFE dataset containing 162 one-minute sequences using simulator, and capture a real-world testset using our custom-built sparse LFE acquisition system, covering diverse indoor and outdoor scenes. Extensive experiments demonstrate that S2D-LFE successfully generates up to 9x9 dense LFE from sparse 2x2 inputs and outperforms existing methods on both synthetic and real-world data. The datasets and code are available at https://github.com/Yutong2022/S2D-LFE.

Self-supervised image denoising methods have garnered significant research attention in recent years, for this kind of method reduces the requirement of large training datasets.Compared to supervised methods, self-supervised methods rely more on the prior embedded in deep networks themselves. As a result, most of the self-supervised methods are designed with Convolution Neural Networks(CNNs) architectures, which well capture one of the most important image prior, translation equivariant prior. Inspired by the great success achieved by the introduction of translational equivariance, in this paper, we explore the way to further incorporate another important image prior. Specifically, we first apply high-accuracy rotation equivariant convolution to self-supervised image denoising. Through rigorous theoretical analysis, we have proved that simply replacing all the convolution layers with rotation equivariant convolution layers would modify the network into its rotation equivariant version.To the best of our knowledge, this is the first time that rotation equivariant image prior is introduced to self-supervised image denoising at the network architecture level with a comprehensive theoretical analysis of equivariance errors, whichoffers a new perspective to the field of self-supervised image denoising.Moreover, to further improve the performance, we design a new mask mechanism to fusion the output of rotation equivariant network and vanilla CNN-based network, and construct an adaptive rotation equivariant framework. Through extensive experiments on three typical methods, we have demonstrated the effectiveness of the proposed method.

Indoor scene texture synthesis has garnered significant interest due to its important potential applications in virtual reality, digital media and creative arts. Existing diffusion-model-based researches either rely on per-view inpainting techniques, which are plagued by severe cross-view inconsistencies and conspicuous seams, or adopt optimization-based approaches that involve substantial computational overhead. In this work, we present **RoomPainter**, a framework that seamlessly integrates efficiency and consistency to achieve high-fidelity texturing of indoor scenes. The core of RoomPainter features a zero-shot technique that effectively adapts a 2D diffusion model for 3D-consistent texture synthesis, along with a two-stage generation strategy that ensures both global and local consistency. Specifically, we introduce Attention-Guided Multi-View Integrated Sampling (**MVIS**) combined with a neighbor-integrated attention mechanism for zero-shot texture map generation. Using the **MVIS**, we firstly generate texture map for the entire room to ensure global consistency, then adopt its variant, namely Attention-Guided Multi-View Integrated Repaint Sampling (**MVRS**) to repaint individual instances within the room, thereby further enhancing local consistency and addressing the occlusion problem. Experiments demonstrate that RoomPainter achieves superior performance for indoor scene texture synthesis in visual quality, global consistency and generation efficiency.

We address the challenge of generating dynamic 4D scenes from monocular multi-object videos with heavy occlusions and introduce Robust4DGen, a novel approach that integrates rendering-based deformable 3D Gaussian optimization with generative priors for view synthesis. While existing view-synthesis models excel at novel view generation for isolated objects, they struggle with full scenes due to their complexity and data demands. To overcome this, Robust4DGen decomposes scenes into individual objects, optimizing a differentiable set of deformable Gaussians per object while capturing 2D occlusions from a 3D perspective through joint Gaussian splatting. Joint splatting ensures occlusion-aware rendering losses in observed frames while explicit object decomposition allows the usage of object-centric diffusion models for object completion in unobserved viewpoints. To reconcile the differences between object-centric priors and the global frame-centric coordinate system of the video, Robust4DGen employs differentiable transformations to unify the rendering and generative constraints within a single framework. The result is a model capable of generating 4D objects across space and time while producing 2D and 3D point tracks from monocular videos. To rigorously evaluate the quality of scene generation and the accuracy of the motion under multi-object occlusions, we introduce MOSE-PTS, a subset of the challenging MOSE benchmark, which we annotated with high-quality 2D point tracks. Quantitative evaluations and perceptual human studies confirm that Robust4DGen generates more realistic novel views of scenes and produces more accurate point tracks compared to existing approaches.

Recent monocular 3D shape reconstruction methods have shown promising zero-shot results on object-segmented images without any occlusions. However, their effectiveness is significantly compromised in real-world conditions, due to imperfect object segmentation by off-the-shelf models and the prevalence of occlusions. To effectively address these issues, we propose a unified regression model that integrates segmentation and reconstruction, specifically designed for occlusion-aware 3D shape reconstruction. To facilitate its reconstruction in the wild, we also introduce a scalable data synthesis pipeline that simulates a wide range of variations in objects, occluders, and backgrounds. Training on our synthetic data enables the proposed model to achieve state-of-the-art zero-shot results on real-world images, using significantly fewer parameters than competing approaches.

Image restoration aims to recover high-quality (HQ) images from degraded low-quality (LQ) ones by reversing the effects of degradation. Existing generative models for image restoration, including diffusion and score-based models, often treat the degradation process as a stochastic transformation, which introduces inefficiency and complexity. In this work, we propose ResFlow, a novel image restoration framework that models the degradation process as a deterministic path using continuous normalizing flows. ResFlow augments the degradation process with an auxiliary process that disambiguates the uncertainty in HQ prediction to enable reversible modeling of the degradation process. ResFlow adopts entropy-preserving flow paths and learns the augmented degradation flow by matching the velocity field. ResFlow significantly improves the performance and speed of image restoration, completing the task in fewer than four sampling steps. Extensive experiments demonstrate that ResFlow achieves state-of-the-art results across various image restoration benchmarks, offering a practical and efficient solution for real-world applications.

Recent deep-learning-based approaches to single-image reflection removal have shown promising advances, primarily for two reasons: 1) the utilization of recognition-pretrained features as inputs, and 2) the design of dual-stream interaction networks. However, according to the Information Bottleneck principle, high-level semantic clues tend to be compressed or discarded during layer-by-layer propagation. Additionally, interactions in dual-stream networks follow a fixed pattern across different layers, limiting overall performance. To address these limitations, we propose a novel architecture called Reversible Decoupling Network (RDNet), which employs a reversible encoder to secure valuable information while flexibly decoupling transmission- and reflection-relevant features during the forward pass. Furthermore, we customize a transmission-rate-aware prompt generator to dynamically calibrate features, further boosting performance. Extensive experiments demonstrate the superiority of RDNet over existing SOTA methods on five widely-adopted benchmark datasets. Our code will be made publicly available.