CVPR 2025 Papers

We introduce SemanticDraw, a new paradigm of interactive content creation where high-quality images are generated in near real-time from given multiple hand-drawn regions, each encoding prescribed semantic meaning. In order to maximize the productivity of content creators and to fully realize their artistic imagination, it requires both quick interactive interfaces and fine-grained regional controls in their tools. Despite astonishing generation quality from recent diffusion models, we find that existing approaches for regional controllability are very slow (52 seconds for 512 x 512 image) while not compatible with acceleration methods such as LCM, blocking their huge potential in interactive content creation. From this observation, we build our solution for interactive content creation in two steps: (1) we establish compatibility between region-based controls and acceleration techniques for diffusion models, maintaining high fidelity of multi-prompt image generation with x 10 reduced number of inference steps, (2) we increase the generation throughput with our new multi-prompt stream batch pipeline, enabling low-latency generation from multiple, region-based text prompts on a single RTX 2080 Ti GPU. Our proposed framework is generalizable to any existing diffusion models and acceleration schedulers, allowing sub-second (0.64 seconds) image content creation application upon well-established image diffusion models.

Most vision-language models today are primarily trained on English image-text pairs, with non-English pairs often filtered out. Evidence from cross-cultural psychology suggests that this approach will bias models against perceptual modes exhibited by people who speak other (non-English) languages. We investigate semantic and expressive variation in image captions across different languages; we analyze both human-annotated datasets and model-produced captions. By analyzing captions across seven languages (English, French, German, Russian, Chinese, Japanese, Korean) in high-quality image captioning datasets (Crossmodal and Visual Genome), we find that multilingual caption sets tend to provide richer visual descriptions than monolingual (including English-only) ones; multilingual sets contain 46.0% more objects, 66.1% more relationships, and 66.8% more attributes. We observe the same results with multilingual captions produced by LLaVA and the Google Vertex API: for example, compared to monolingual captions, they cover 21.9% more objects,18.8% more relations, and 20.1% more attributes. These suggest that, across a large number of samples, different languages bias people and models to focus on different visual concepts. Finally, we show that models trained on image-text data in one language perform distinctly better on that language's test set. Our work points towards the potential value of training vision models on multilingual data sources to widen the range/variation of descriptive information those models are exposed to.

Generating reasonable and high-quality human interactive motions in a given dynamic environment is crucial for understanding, modeling, transferring, and applying human behaviors to both virtual and physical robots. In this paper, we introduce an effective method, SemGeoMo, for dynamic contextual human motion generation, which fully leverages the text-affordance-joint multi-level semantic and geometric guidance in the generation process, improving the semantic rationality and geometric correctness of generative motions. Our method achieves state-of-the-art performance on three datasets and demonstrates superior generalization capability for diverse interaction scenarios.

Diffusion models achieved unprecedented fidelity and diversity for synthesizing image, video, 3D assets, etc. However, subject mixing is an unresolved issue for diffusion-based image synthesis, particularly for synthesizing multiple similar-looking subjects. We propose Self-Cross Diffusion Guidance to penalize the overlap between cross-attention maps and the aggregated self-attention map. Compared to previous methods based on self-attention or cross-attention alone, our guidance is more effective in eliminating subject mixing. What's more, our guidance addresses subject mixing for all relevant patches beyond the most discriminant one, e.g., the beak of a bird. For each subject, we aggregate self-attention maps of patches with higher cross-attention values. Thus, the aggregated self-attention map forms a region that the whole subject attends to. Our training-free method boosts the performance of both Unet-based and Transformer-based diffusion models such as the Stable Diffusion series. We also release a similar subjects dataset (SSD), a challenging benchmark, and utilize GPT-4o for automatic and reliable evaluation. Extensive qualitative and quantitative results demonstrate the effectiveness of our self-cross diffusion guidance.

Multimodal sentiment analysis has attracted extensive research attention as increasing users share images and texts to express their emotions and opinions on social media. Collecting large amounts of labeled sentiment data is an expensive and challenging task due to the high cost of labeling and unavoidable label ambiguity. Semi-supervised learning (SSL) is explored to utilize the extensive unlabeled data to alleviate the demand for annotation. However, different from typical multimodal tasks, the inconsistent sentiment between image and text leads to the sub-optimal performance of SSL algorithms. To address this issue, we propose SCDR, the first semi-supervised image-text sentiment recognition framework. To better utilize the discriminative features of each modality, we decouple features into common and private parts and then use the private features to train unimodal classifiers for enhanced modality-specific sentiment representation. Considering the complex relation between modalities, we devise a modal selection-based attention module that adaptively assesses the dominant sentiment modality at the sample level to guide the fusion of multimodal representations. Furthermore, to prevent the model predictions from overly relying on common features under the guidance of multimodal labels, we design a pseudo-label filtering strategy based on the matching degree of prediction and dominant modality. Extensive experiments and comparisons on five publicly available datasets demonstrate that SCDR outperforms state-of-the-art methods. The code is provided in the supplementary material and will be released to the public.

The CLIP model has demonstrated significant advancements in aligning visual and language modalities through large-scale pre-training on image-text pairs, enabling strong zero-shot classification and retrieval capabilities on various domains. However, CLIP's training remains computationally intensive, with high demands on both data processing and memory. To address these challenges, recent masking strategies have emerged, focusing on the selective removal of image patches to improve training efficiency. Although effective, these methods often compromise key semantic information, resulting in suboptimal alignment between visual features and text descriptions.In this work, we present a concise yet effective approach called Patch Generation-to-Selection (CLIP-PGS) to enhance CLIP's training efficiency while preserving critical semantic content. Our method introduces a gradual masking process in which a small set of candidate patches is first pre-selected as potential mask regions. Then, we apply Sobel edge detection across the entire image to generate an edge mask that prioritizes the retention of the primary object areas. Finally, similarity scores between the candidate mask patches and their neighboring patches are computed, with optimal transport normalization refining the selection process to ensure a balanced similarity matrix.Our approach, CLIP-PGS, sets new state-of-the-art results in zero-shot classification and retrieval tasks, achieving superior performance in robustness evaluation and language compositionality benchmarks.

Significant advances have been made in the sampling efficiency of diffusion and flow matching models, driven by Consistency Distillation (CD), which trains a student model to mimic the output of a teacher model at a later timestep. However, we found that the knowledge discrepancy between student and teacher varies significantly across different timesteps, leading to suboptimal performance in CD.To address this issue, we propose the Curriculum Consistency Model (CCM), which stabilizes and balances the knowledge discrepancy across timesteps. Specifically, we regard the distillation process at each timestep as a curriculum and introduce a metric based on the Peak Signal-to-Noise Ratio (PSNR) to quantify the knowledge discrepancy of this curriculum, then ensure that the curriculum maintains consistent knowledge discrepancy across different timesteps by having the teacher model iterate more steps when the noise intensity is low.Our method achieves competitive single-step sampling Frechet Inception Distance (FID) scores of 1.64 on CIFAR-10 and 2.18 on ImageNet 64x64.Moreover, we have extended our method to large-scale text-to-image models and confirmed that it generalizes well to both diffusion models (Stable Diffusion XL) and flow matching models (Stable Diffusion 3). The generated samples demonstrate improved image-text alignment and semantic structure since CCM enlarges the distillation step at large timesteps and reduces the accumulated error.

In this paper, we introduce SearchDet, a training-free long-tail object detection framework that significantly enhances open-vocabulary object detection performance. SearchDet retrieves a set of positive and negative images of an object to ground, embeds these images, and computes an input image--weighted query which is used to detect the desired concept in the image. Our proposed method is simple and training-free, yet achieves over 16.81% mAP improvement on ODinW and 59.85% mAP improvement on LVIS compared to state-of-the-art models such as GroundingDINO. We further show that our approach of basing object detection on a set of Web-retrieved exemplars is stable with respect to variations in the exemplars, suggesting a path towards eliminating costly data annotation and training procedures.

Image-based relighting of indoor rooms creates an immersive virtual understanding of the space, which is useful for interior design, virtual staging, and real estate. Relighting indoor rooms from a single image is especially challenging due to complex illumination interactions between multiple lights and cluttered objects featuring a large variety in geometrical and material complexity. Recently, generative models have been successfully applied to image-based relighting conditioned on a target image or a latent code, albeit without detailed local lighting control. In this paper, we introduce ScribbleLight, a generative model that supports local fine-grained control of lighting effects through scribbles that describe changes in lighting. Our key technical novelty is an Albedo-conditioned Stable Image Diffusion model that preserves the intrinsic color and texture of the original image after relighting and an encoder-decoder-based ControlNet architecture that enables geometry-preserving lighting effects with normal map and scribble annotations. We demonstrate ScribbleLight's ability to create different lighting effects (e.g., turning lights on/off, adding highlights, cast shadows, or indirect lighting from unseen lights) from sparse scribble annotations.

We present a novel approach to integrating scientific knowledge into generative models, enhancing their realism and consistency in image synthesis. First, we introduce Science-T2I, an expert-annotated adversarial dataset comprising adversarial 20k image pairs with 9k prompts, covering wide distinct scientific knowledge categories. Leveraging Science-T2I, we present SciScore, an end-to-end reward model that refines the assessment of generated images based on scientific knowledge, which is achieved by augmenting both the scientific comprehension and visual capabilities of pre-trained CLIP model. Additionally, based on Science-T2I, we propose a two-stage training framework, comprising a supervised fine-tuning phase and a masked online fine-tuning phase, to incorporate scientific knowledge into existing generative models. Through comprehensive experiments, we demonstrate the effectiveness of our framework in establishing new standards for evaluating the scientific realism of generated content. Specifically, SciScore attains performance comparable to human-level, demonstrating a 5% improvement similar to evaluations conducted by experienced human evaluators. Furthermore, by applying our proposed fine-tuning method to FLUX, we achieve a performance enhancement exceeding 50% on SciScore.

We present SceneFactor, a diffusion-based approach for large-scale 3D scene generation that enables controllable generation and effortless editing. SceneFactor enables text-guided 3D scene synthesis through our factored diffusion formulation, leveraging latent semantic and geometric manifolds for generation of arbitrary-sized 3D scenes. While text input enables easy, controllable generation, text guidance remains imprecise for intuitive, localized editing and manipulation of the generated 3D scenes. Our factored semantic diffusion generates a proxy semantic space composed of semantic 3D boxes that enables controllable editing of generated scenes by adding, removing, changing the size of the semantic 3D proxy boxes that guides high-fidelity, consistent 3D geometric editing. Extensive experiments demonstrate that our approach enables high-fidelity 3D scene synthesis with effective controllable editing through our factored diffusion approach.

The reconstruction of immersive and realistic 3D scenes holds significant practical importance in various fields of computer vision and computer graphics. Typically, immersive and realistic scenes should be free from obstructions by dynamic objects, maintain global texture consistency, and allow for unrestricted exploration. The current mainstream methods for image-driven scene construction involves iteratively refining the initial image using a moving virtual camera to generate the scene. However, previous methods struggle with visual discontinuities due to global texture inconsistencies under varying camera poses, and they frequently exhibit scene voids caused by foreground-background occlusions. To this end, we propose a novel layered 3D scene reconstruction framework from panoramic image, named Scene4U. Specifically, Scene4U integrates an open-vocabulary segmentation model with a large language model to decompose a real panorama into multiple layers. Then, we employs a layered repair module based on diffusion model to restore occluded regions using visual cues and depth information, generating a hierarchical representation of the scene. The multi-layer panorama is then initialized as a 3D Gaussian Splatting representation, followed by layered optimization, which ultimately produces an immersive 3D scene with semantic and structural consistency that supports free exploration. Our Scene4U outperforms state-of-the-art method, improving by 24.24% in LPIPS and 24.40% in BRISQUE, while also achieving the fastest training speed. Additionally, to demonstrate the robustness of Scene4U and allow users to experience immersive scenes from various landmarks, we build WorldVista3D dataset for 3D scene reconstruction, which contains panoramic images of globally renowned sites. The implementation code and dataset will be made publicly available.

In this paper, we propose Scene Splatter, a momentum-based paradigm for video diffusion to generate generic scenes from single image. Existing methods, which employ video generation models to synthesize novel views, suffer from limited video length and scene inconsistency, leading to artifacts and distortions during further reconstruction. To address this issue, we construct noisy samples from original features as momentum to enhance video details and maintain scene consistency. However, for latent features with the perception field that spans both known and unknown regions, such latent-level momentum restricts the generative ability of video diffusion in unknown regions. Therefore, we further introduce the aforementioned consistent video as a pixel-level momentum to a directly generated video without momentum for better recovery of unseen regions. Our cascaded momentum enables video diffusion models to generate both high-fidelity and consistent novel views. We further finetune the global Gaussian representations with enhanced frames and render new frames for momentum update in the next step. In this manner, we can iteratively recover a 3D scene, avoiding the limitation of video length. Extensive experiments demonstrate the generalization capability and superior performance of our method in high-fidelity and consistent scene generation.

Navigation instruction generation (NIG), which provides interactive feedback and guidance to humans along a trajectory, is vital for developing embodied agents capable of human-machine communication and collaboration through natural language. Early data-driven methods directly map sequences of past observations to trajectory descriptions on limited datasets, lacking the necessary spatial understanding in complex 3D environments. While recent approaches leverage Large Language Models (LLMs) to improve NIG, they often overlook the global spatial context in navigation, such as the inherent space discretization in maps. Instead of straightforwardly feeding textual descriptions of the map into LLMs, we propose a scene map-based prompt tuning framework for NIG, MAPInstructor, which incorporates map context for parameter-efficient updating of LLMs. MAPInstructor comprises three key components: (i) scene representation encoding, where egocentric observations are projected into 3D voxels for fine-grained scene understanding; (ii) map prompt tuning, which integrates a topological map representation of the entire trajectory into an LLM-based decoder; and (iii) landmark uncertainty assessment, which mitigates hallucinations in landmark predictions, thereby enhancing the reliability and coherence of instruction generation. Extensive experiments on three navigation datasets (i.e., R2R, REVERIE, RxR) confirm the generalization and effectiveness of our algorithm.

Traditional segmentation models, while effective in isolated tasks, often fail to generalize to more complex and open-ended segmentation problems, such as free-form, open-vocabulary, and in-the-wild scenarios. To bridge this gap, we propose to scale up image segmentation across diverse datasets and tasks such that the knowledge across different tasks and datasets can be integrated while improving the generalization ability. QueryMeldNet, a novel segmentation framework, is introduced and designed to scale seamlessly across both data size and task diversity. It is built upon a dynamic object query mechanism called query meld, which fuses different types of queries using cross-attention. This hybrid approach enables the model to balance between instance- and stuff-level segmentation, providing enhanced scalability for handling diverse object types. We further enhance scalability by leveraging synthetic data-generating segmentation masks and captions for pixel-level and open-vocabulary tasks-drastically reducing the need for costly human annotations. By training on multiple datasets and tasks at scale, QueryMeldNet continuously improves performance as the volume and diversity of data and tasks increase. It exhibits strong generalization capabilities, boosting performance in open-set segmentation tasks SeginW by 7 points. These advancements mark a key step toward universal, scalable segmentation models capable of addressing the demands of real-world applications.

We propose a compressive yet effective mesh tokenization, Blocked and Patchified Tokenization (BPT), facilitating the generation of meshes exceeding 8k faces. BPT compresses mesh sequences by employing block-wise indexing and patch aggregation, reducing their length by approximately 75% compared to the vanilla sequences. This compression milestone unlocks the potential to utilize mesh data with significantly more faces, thereby enhancing detail richness and improving generation robustness. Empowered with the BPT, we have built a foundation mesh generative model training on scaled mesh data to support flexible control for point clouds and images. Our model demonstrates the capability to generate meshes with intricate details and accurate topology, achieving SoTA performance on mesh generation and reaching the level for direct product usage.

Text encoders in diffusion models have rapidly evolved, transitioning from CLIP to T5-XXL. Although this evolution has significantly enhanced the models' ability to understand complex prompts and generate text, it also leads to a substantial increase in the number of parameters. Despite T5 series encoders being trained on the C4 natural language corpus, which includes a significant amount of non-visual data, diffusion models with T5 encoder do not respond to those non-visual prompts, indicating redundancy in representational power. Therefore, it raises an important question: "Do we really need such a large text encoder?" In pursuit of an answer, we employ vision-based knowledge distillation to train a series of T5 encoder models. To fully inherit its capabilities, we constructed our dataset based on three criteria: image quality, semantic understanding, and text-rendering. Our results demonstrate the scaling down pattern that the distilled T5-base model can generate images of comparable quality to those produced by T5-XXL, while being 50 times smaller in size. This reduction in model size significantly lowers the GPU requirements for running state-of-the-art models such as FLUX and SD3, making high-quality text-to-image generation more accessible.

The scaling law has been validated in various domains, such as natural language processing (NLP) and massive computer vision tasks; however, its application to motion generation remains largely unexplored. In this paper, we introduce a scalable motion generation framework that includes the motion tokenizer Motion FSQ-VAE and a text-prefix autoregressive transformer. Through comprehensive experiments, we observe the scaling behavior of this system. For the first time, we confirm the existence of scaling laws within the context of motion generation. Specifically, our results demonstrate that the normalized test loss of our prefix autoregressive models adheres to a logarithmic law in relation to compute budgets. Furthermore, we also confirm the power law between Non-Vocabulary Parameters, Vocabulary Parameters, and Data Tokens with respect to compute budgets respectively. Leveraging the scaling law, we predict the optimal transformer size, vocabulary size, and data requirements for a compute budget of 1e18. The test loss of the system, when trained with the optimal model size, vocabulary size, and required data, aligns precisely with the predicted test loss.