CVPR 2025 Papers

We introduce PersonaHOI, a training- and tuning-free framework that fuses a general StableDiffusion model with a personalized face diffusion (PFD) model to generate identity-consistent human-object interaction (HOI) images. While existing PFD models have advanced significantly, they often overemphasize facial features at the expense of full-body coherence, PersonaHOI introduces an additional StableDiffusion (SD) branch guided by HOI-oriented text inputs. By incorporating cross-attention constraints in the PFD branch and spatial merging at both latent and residual levels, PersonaHOI preserves personalized facial details while ensuring interactive non-facial regions. Experiments, validated by a novel interaction alignment metric, demonstrate the superior realism and scalability of PersonaHOI, establishing a new standard for practical personalized face with HOI generation. Code is available at https://github.com/JoyHuYY1412/PersonaHOI.

Finetuning-free personalized image generation can synthesize customized images without test-time finetuning, attracting wide research interest owing to its high efficiency. Current finetuning-free methods simply adopt a single training stage with a simple image reconstruction task, and they typically generate low-quality images inconsistent with the reference images during test-time. To mitigate this problem, inspired by the recent DPO (i.e., direct preference optimization) technique, this work proposes an additional training stage to improve the pre-trained personalized generation models. However, traditional DPO only determines the overall superiority or inferiority of two samples, which is not suitable for personalized image generation because the generated images are commonly inconsistent with the reference images only in some local image patches. To tackle this problem, this work proposes PatchDPO that estimates the quality of image patches within each generated image and accordingly trains the model. To this end, PatchDPO first leverages the pre-trained vision models with a proposed self-supervised training method to estimate the patch quality. Next, PatchDPO adopts a weighted training approach to train the model with the estimated patch quality, which rewards the image patches with high quality while penalizing the image patches with low quality. Experiment results demonstrate that PatchDPO significantly improves the performance of multiple pre-trained personalized generation models, and achieves state-of-the-art performance on both single-object and multi-object personalized image generation. Our code is available at https://github.com/hqhQAQ/PatchDPO.

Existing image-to-3D creation methods typically split the task into two individual stage: multi-view image generation and 3D reconstruction, leading to two main limitations: (1) In multi-view generation stage, the multi-view generated images present a challenge to preserving 3D consistency;; (2) In 3D reconstruction stage, there is a domain gap between real training data and generated multi-view input during inference. To address these issues, we propose Ouroboros3D, end-to-end trainable framework that integrates multi-view generation and 3D reconstruction into a recursive diffusion process through feedback mechanism.Our framework operates through iterative cycles where each cycle consists of a denoising process and a reconstruction step.By incorporating a 3D-aware feedback mechanism, our multi-view generative model leverages the explicit 3D geometric information (e.g. texture, position) from the feedback of reconstruction results of the previous process as conditions, thus modeling consistency at the 3D geometric level. Furthermore, through joint training of both the multi-view generative and reconstruction models, we alleviate reconstruction stage domain gap and enable mutual enhancement within the recursive process. Experimental results demonstrate that Ouroboros3D outperforms methods that treat these stages separately and those that combine them only during inference, achieving superior multi-view consistency and producing 3D models with higher geometric realism.

Transformers today still struggle to generate one-minute videos because self-attention layers are inefficient for long context. Alternatives such as Mamba layers struggle to produce coherent scenes because their hidden states are small and less expressive. We experiment with Test-Time Training (TTT) layers, whose hidden states themselves can be neural networks, therefore larger and more expressive. Adding TTT layers into a pre-trained Transformer enables it to generate one-minute videos from text storyboards. We curate a dataset based on Tom and Jerry cartoons as a proof-of-concept benchmark. Compared to baselines such as Mamba 2, Gated DeltaNet, and sliding-window attention layers, TTT layers generate much more coherent videos that tell complete stories, leading by 34 Elo points in a human evaluation of 100 videos per method. Although promising, our results are still limited in physical realism, and the efficiency of our implementation can be further improved. Sample videos, code and annotations are available at: https://test-time-training.github.io/video-dit

The diffusion models, in early stages focus on constructing basic image structures, while the refined details, including local features and textures, are generated in later stages. Thus the same network layers are forced to learn both structural and textural information simultaneously, significantly differing from the traditional deep learning architectures (e.g., ResNet or GANs) which captures or generates the image semantic information at different layers. This difference inspires us to explore the time-wise diffusion models. We initially investigate the key contributions of the U-Net parameters to the denoising process and identify that properly zeroing out certain parameters (including large parameters) contributes to denoising, substantially improving the generation quality on the fly. Capitalizing on this discovery, we propose a simple yet effective method--termed "MaskUNet"-- that enhances generation quality with negligible parameter numbers. Our method fully leverages timestep- and sample-dependent effective U-Net parameters. To optimize MaskUNet, we offer two fine-tuning strategies: a training-based approach and a training-free approach, including tailored networks and optimization functions. In zero-shot inference on the COCO dataset, MaskUNet achieves the best FID score and further demonstrates its effectiveness in downstream task evaluations. Project page: https://gudaochangsheng.github.io/MaskUnet-Page/

In training-free conditional generative tasks, diffusion models utilize differentiable loss functions to steer the generative reverse process, necessitating modifications to sampling algorithms like DDPM and DDIM. However, such adjustments likely reduce flexibility and reliability. In this paper, we propose NoiseCtrl, a sampling-algorithm-agnostic technique for controlled image generation. Essentially, diffusion models generate denoised results z_ t-1 by adding a predicted mean \mu_ t with random noise epsilon_ t . NoiseCtrl specifically adjusts the random noise while leaving the underlying sampling algorithms unchanged. At each step t , NoiseCtrl converts the unconditional Gaussian noise into conditional noise epsilon'_ t by substituting the isotropic Gaussian distribution with the von Mises-Fisher distribution. This substitution introduces a directional focus while preserving the randomness required for conditional image generation. Thanks to this non-intrusive design, NoiseCtrl is straightforward to integrate and has been extensively validated through experiments, demonstrating its adaptability for different diffusion algorithms and superior performance across various conditional generation tasks.

In scene graph generation (SGG), the accurate prediction of unseen triples is essential for its effectiveness in downstream vision-language tasks. We hypothesize that the predicates of unseen triples can be viewed as transformations of seen predicates in feature space, and the essence of the zero-shot task is to bridge the gap caused by this transformation. Traditional models, however, have difficulty addressing this challenge, which we attribute to their inability to model the predicates equivariant. To overcome this limitation, we introduce a novel framework based on capsule networks (CAPSGG). We propose a Three-Stream Pipeline that generates modality-specific representations for predicates, while building low-level predicate capsules of these modalities. Then these capsules are aggregated into high-level predicate capsules using a Routing Capsule Layer. In addition, we introduce GroupLoss to aggregate capsules with the same predicate label into groups. This replaces the global loss with the intra-group loss, effectively balancing the learning of predicate invariance and equivariant features, while mitigating the impact of the severe long-tail distribution of the predicate categories. Our extensive experiments demonstrate the notable superiority of our approach over state-of-the-art methods, with zero-shot indicators outperforming up to 132.26% on SGCls task than the T-CAR [21]. Our code will be available upon publication.

Generating realistic human videos remains a challenging task, with the most effective methods currently relying on a human motion sequence as a control signal. Existing approaches often use existing motion extracted from other videos, which restricts applications to specific motion types and global scene matching. We propose Move-in-2D, a novel approach to generate human motion sequences conditioned on a scene image, allowing for diverse motion that adapts to different scenes. Our approach utilizes a diffusion model that accepts both a scene image and text prompt as inputs, producing a motion sequence tailored to the scene. To train this model, we collect a large-scale video dataset featuring single-human activities, annotating each video with the corresponding human motion as the target output. Experiments demonstrate that our method effectively predicts human motion that aligns with the scene image after projection. Furthermore, we show that the generated motion sequence improves human motion quality in video synthesis tasks.

Masked Image Modeling (MIM) with Vector Quantization (VQ) has achieved great success in both self-supervised pre-training and image generation. However, most existing methods struggle to address the trade-off in the shared latent space for generation quality vs. representation learning and efficiency. To push the limits of this paradigm, we propose MergeVQ, which incorporates token merging techniques into VQ-based generative models to bridge the gap between image generation and visual representation learning in a unified architecture. During pre-training, MergeVQ decouples top-k semantics from latent space with the token merge module after self-attention blocks in the encoder for subsequent Look-up Free Quantization (LFQ) and global alignment and recovers their fine-grained details through cross-attention in the decoder for reconstruction. As for second-stage generation, we introduce MergeAR, which performs KV Cache compression for efficient raster-order prediction. Extensive experiments on ImageNet verify that MergeVQ as an AR generative model achieves competitive performance in both visual representation learning and image generation tasks while maintaining favorable token efficiency and inference speed. The code and model will be available at https://apexgen-x.github.io/MergeVQ.

Pre-training with random masked inputs has emerged as a novel trend in self-supervised training. However, supervised learning still faces a challenge in adopting masking augmentations, primarily due to unstable training. In this paper, we propose a novel way to involve masking augmentations dubbed Masked Sub-branch (MaskSub). MaskSub consists of the main-branch and sub-branch, the latter being a part of the former. The main-branch undergoes conventional training recipes, while the sub-branch merits intensive masking augmentations, during training. MaskSub tackles the challenge by mitigating adverse effects through a relaxed loss function similar to a self-distillation loss. Our analysis shows that MaskSub improves performance, with the training loss converging faster than in standard training, which suggests our method stabilizes the training process. We further validate MaskSub across diverse training scenarios and models, including DeiT-III training, MAE finetuning, CLIP finetuning, BERT training, and hierarchical architectures (ResNet and Swin Transformer). Our results show that MaskSub consistently achieves impressive performance gains across all the cases. MaskSub provides a practical and effective solution for introducing additional regularization under various training recipes. Code available at https://github.com/naver-ai/augsub

We introduce MVGenMaster, a multi-view diffusion model enhanced with 3D priors to address versatile Novel View Synthesis (NVS) tasks. MVGenMaster leverages 3D priors that are warped using metric depth and camera poses, significantly enhancing both generalization and 3D consistency in NVS. Our model features a simple yet effective pipeline that can generate up to 100 novel views conditioned on variable reference views and camera poses with a single forward process. Additionally, we have developed a comprehensive large-scale multi-view image dataset called MvD-1M, comprising up to 1.6 million scenes, equipped with well-aligned metric depth to train MVGenMaster. Moreover, we present several training and model modifications to strengthen the model with scaled-up datasets. Extensive evaluations across in- and out-of-domain benchmarks demonstrate the effectiveness of our proposed method and data formulation.

Recent sparse multi-view scene reconstruction advances like DUSt3R and MASt3R no longer require camera calibration and camera pose estimation. However, they only process a pair of views at a time to infer pixel-aligned pointmaps. When dealing with more than two views, a combinatorial number of error prone pairwise reconstructions are usually followed by an expensive global optimization, which often fails to rectify the pairwise reconstruction errors. To handle more views, reduce errors, and improve inference time, we propose the fast single-stage feed-forward network MV-DUSt3R. At its core are multi-view decoder blocks which exchange information across any number of views while considering one reference view. To make our method robust to reference view selection, we further propose MV-DUSt3R+, which employs cross-reference-view blocks to fuse information across different reference view choices. To further enable novel view synthesis, we extend both by adding and jointly training Gaussian splatting heads. Experiments on multi-view stereo reconstruction, multi-view pose estimation, and novel view synthesis confirm that our methods improve significantly upon prior art. Code released.

Recent motion-aware large language models have demonstrated promising potential in unifying motion comprehension and generation. However, existing approaches primarily focus on coarse-grained motion-text modeling, where text describes the overall semantics of an entire motion sequence in just a few words. This limits their ability to handle fine-grained motion-relevant tasks, such as understanding and controlling the movements of specific body parts. To overcome this limitation, we pioneer MG-MotionLLM, a unified motion-language model for multi-granular motion comprehension and generation. We further introduce a comprehensive multi-granularity training scheme by incorporating a set of novel auxiliary tasks, such as localizing temporal boundaries of motion segments via detailed text as well as motion detailed captioning, to facilitate mutual reinforcement for motion-text modeling across various levels of granularity. Extensive experiments show that our MG-MotionLLM achieves superior performance on classical text-to-motion and motion-to-text tasks, and exhibits potential in novel fine-grained motion comprehension and editing tasks. Project page: CVI-SZU/MG-MotionLLM

Multiview diffusion models have shown considerable success in image-to-3D generation for general objects. However, when applied to human data, existing methods have yet to deliver promising results, largely due to the challenges of scaling multiview attention to higher resolutions. In this paper, we explore human multiview diffusion models at the megapixel level and introduce a solution called "mesh attention" to enable training at 1024x1024 resolution. Using a clothed human mesh as a central coarse geometric representation, the proposed mesh attention leverages rasterization and projection to establish direct cross-view coordinate correspondences. This approach significantly reduces the complexity of multiview attention while maintaining cross-view consistency. Building on this foundation, we devise a mesh attention block and combine it with keypoint conditioning to create our human-specific multiview diffusion model, MEAT. In addition, we present valuable insights into applying multiview human motion videos for diffusion training, addressing the longstanding issue of data scarcity. Extensive experiments show that MEAT effectively generates dense, consistent multiview human images at the megapixel level, outperforming existing multiview diffusion methods. Code is available at https://johann.wang/MEAT/.

Customized text-to-image generation, which synthesizes images based on user-specified concepts, has made significant progress in handling individual concepts. However, when extended to multiple concepts, existing methods often struggle with properly integrating different models and avoiding the unintended blending of characteristics from distinct concepts. In this paper, we propose MC^2, a novel approach for multi-concept customization that enhances flexibility and fidelity through inference-time optimization. MC^2 enables the integration of multiple single-concept models with heterogeneous architectures. By adaptively refining attention weights between visual and textual tokens, our method ensures that image regions accurately correspond to their associated concepts while minimizing interference between concepts. Extensive experiments demonstrate that MC^2 outperforms training-based methods in terms of prompt-reference alignment. Furthermore, MC^2 can be seamlessly applied to text-to-image generation, providing robust compositional capabilities. To facilitate the evaluation of multi-concept customization, we also introduce a new benchmark, MC++. The code is available at https://github.com/JIANGJiaXiu/MC-2.

We introduce Presto, a novel video diffusion model designed to generate 15-second videos with long-range coherence and rich content. Extending video generation to maintain scenario diversity over long durations presents significant challenges. To address this, we propose a Segmented Cross-Attention (SCA) strategy, which splits hidden states into segments along the temporal dimension, allowing each segment to cross-attend to a corresponding sub-caption. SCA requires no additional parameters, enabling seamless incorporation into current DiT-based architectures. To facilitate high-quality long video generation, we build the LongTake-HD dataset, consisting of 261k content-rich videos with scenario coherence, annotated with an overall video caption and five progressive sub-captions. Experiments show that our Presto achieves 78.5% on the VBench Semantic Score and 100% on the Dynamic Degree, outperforming existing state-of-the-art video generation methods. This demonstrates that our proposed Presto significantly enhances the content richness, maintains long-range coherence, and captures intricate textual details. All code and model weights will be made publicly available.

Polynomial Lyapunov function \mathcal V (x) provides mathematically rigorous that converts stability analysis into efficiently solvable optimization problem. Traditional numerical methods rely on user-defined templates, while emerging neural \mathcal V (x) offer flexibility but exhibit poor generalization yield from naive Square polynomial networks. In this paper, we propose a novel learning-enabled polynomial \mathcal V (x) synthesis approach, where a data-driven machine learning process guided by target-based sampling to fit candidate \mathcal V (x) which naturally compatible with the sum-of-squares (SOS) soundness verification. The framework is structured as an iterative loop between a Learner and a Verifier , where the Learner trains expressive polynomial \mathcal V (x) network via polynomial expansions, while the Verifier encodes learned candidates with SOS constraints to identify a real \mathcal V (x) by solving LMI feasibility test problems. The entire procedure is driven by a high-accuracy counterexample guidance technique to further enhance efficiency. Experimental results demonstrate that our approach outperforms both SMT-based polynomial neural Lyapunov function synthesis and traditional SOS method.

Covariance matrices have proven highly effective across many scientific fields. Since these matrices lie within the Symmetric Positive Definite (SPD) manifold--a Riemannian space with intrinsic non-Euclidean geometry, the primary challenge in representation learning is to respect this underlying geometric structure. Drawing inspiration from the success of Euclidean deep learning, researchers have developed neural networks on the SPD manifolds for more faithful covariance embedding learning. A notable advancement in this area is the implementation of Riemannian batch normalization (RBN), which has been shown to improve the performance of SPD network models. Nonetheless, the Riemannian metric beneath the existing RBN might fail to effectively deal with the ill-conditioned SPD matrices (ICSM), undermining the effectiveness of RBN. In contrast, the Bures-Wasserstein metric (BWM) demonstrates superior performance for ill-conditioning. In addition, the recently introduced Generalized BWM (GBWM) parameterizes the vanilla BWM via an SPD matrix, allowing for a more nuanced representation of vibrant geometries of the SPD manifold. Therefore, we propose a novel RBN algorithm based on the GBW geometry, incorporating a learnable metric parameter. Moreover, the deformation of GBWM by matrix power is also introduced to further enhance the representational capacity of GBWM-based RBN. Experimental results on different datasets validate the effectiveness of our proposed method. The code is available at https://github.com/jjscc/GBWBN.

We present a novel approach to enhance camera pose estimation in global Structure-from-Motion (SfM) frameworks by filtering inaccurate pose graph edges - representing relative translation estimates - before applying translation averaging. In SfM, pose graph vertices represent images, and edges represent relative poses (rotations and translations) between cameras. We reformulate the edge filtering problem as a vertex filtering in the dual graph, specifically, a line graph where vertices correspond to edges in the original graph and edges correspond to cameras. Utilizing this representation, we frame the problem as a binary classification over nodes in the dual graph. To identify outlier edges, we employ a Transformer-based architecture. To overcome the challenge of memory overflow caused by converting to a line graph, we introduce a clustering-based graph processing approach, enabling our method to be applied to arbitrarily large pose graphs. Our method outperforms existing relative translation filtering techniques in terms of camera position accuracy and can be seamlessly integrated with other filters. The code is available at https://github.com/DmblnNicole/LFOE-GlobalSfM.