ICCV 2025 Papers

The rapid growth of text-to-image diffusion models has raised concerns about their potential misuse in generat- ing harmful or unauthorized contents. To address these issues, several Concept Erasure methods have been pro- posed. However, most of them fail to achieve both robust- ness, i.e., the ability to robustly remove the target concept., and effectiveness, i.e., maintaining image quality. While few recent techniques successfully achieve these goals for NSFW concepts, none could handle narrow concepts such as copyrighted characters or celebrities. Erasing these nar- row concepts is critical in addressing copyright and legal concerns. However, erasing them is challenging due to their close distances to non-target neighboring concepts, requir- ing finer-grained manipulation. In this paper, we introduce Subspace Mapping (SuMa), a novel method specifically de- signed to achieve both robustness and effectiveness in eas- ing these narrow concepts. SuMa first derives a target sub- space representing the concept to be erased and then neu- tralizes it by mapping it to a reference subspace that mini- mizes the distance between the two. This mapping ensures the target concept is robustly erased while preserving im- age quality. We conduct extensive experiments with SuMa across four tasks: subclass erasure, celebrity erasure, artis- tic style erasure, and instance erasure and compare the results with current state-of-the-art methods. Our method achieves image quality comparable to approaches focused on effectiveness, while also yielding results that are on par with methods targeting completeness.

Stylized face recognition is the task of recognizing generated faces with the same ID across diverse stylistic domains (e.g., anime, painting, cyberpunk styles). This emerging field plays a vital role in the governance of generative image, serving the primary objective: Recognize the ID information of stylized faces to detect potential infringements of portrait rights. Despite its importance, progress in stylized face recognition has been hindered by the lack of large-scale, stylistically diverse datasets. To address this gap, we introduce the Stylized-Face dataset, which is the first dataset specifically designed for stylized face recognition. Stylized-Face dataset includes 4.6 million images across 62k IDs, specifically curated to enhance model performance in stylized face recognition tasks. To ensure data quality (i.e., ID preservation) at this massive scale, we implement a semi-automated pipeline for large-scale data cleaning. Based on the Stylized-Face dataset, we establish three benchmarks to evaluate the robustness and generalization of recognition models across various scenarios, including within-distribution performance, cross-prompt generalization, and cross-method generalization, which target key challenges in stylized face recognition. Experimental results demonstrate that models trained on Stylized-Face achieve remarkable improvements in both stylized face recognition performance (a 15.9% improvement in TAR at FAR=1e-4) and generalization (a 13.3% improvement in TAR at FAR=1e-3 in cross-method generalization).

Scene text image super-resolution (STISR) focuses on enhancing the clarity and readability of low-resolution text images. Existing methods often rely on text probability distribution priors derived from text recognizers to guide the super-resolution process. While effective in capturing general structural information of text, these priors lack the ability to preserve specific text style details, such as font, stereoscopic effect and spatial transformation, leading to a loss of visual quality and stylistic consistency in the super-resolved images. To address these limitations, we propose a Style embedding-based scene text image Super-Resolution Network (StyleSRN), which introduces a text style embedding mechanism to preserve and enhance text style features during the super-resolution process. The proposed architecture includes a Style Enhancement Block for capturing multi-scale cross-channel dependencies, and a Style Content Fusion Block that effectively integrates text content with style information, ensuring that the structure and style of the restored text are not distorted. Furthermore, we introduce a Text Style Loss based on the Gram matrix to supervise the reconstruction process at the style level, thereby maintaining the stylistic consistency of the restored text images. Extensive experiments on the TextZoom dataset and five scene text recognition benchmarks demonstrate the superiority of our method. The code will be released in the future.

We present StyleMotif, a novel Stylized Motion Latent Diffusion model, generating motion conditioned on both content and style from multiple modalities. Unlike existing approaches that either focus on generating diverse motion content or transferring style from sequences, StyleMotif seamlessly synthesizes motion across a wide range of content while incorporating stylistic cues from multi-modal inputs, including motion, text, image, video, and audio. To achieve this, we introduce a style-content cross fusion mechanism and align a style encoder with a pre-trained multi-modal model, ensuring that the generated motion accurately captures the reference style while preserving realism. Extensive experiments demonstrate that our framework surpasses existing methods in stylized motion generation and exhibits emergent capabilities for multi-modal motion stylization, enabling more nuanced motion synthesis. Project Page: https://stylemotif.github.io.

In the domain of text-to-image generation, diffusion models have emerged as powerful tools. Recently, studies on visual prompting, where images are used as prompts, have enabled more precise control over style and content. However, existing methods often suffer from content leakage, where undesired elements of the visual style prompt are transferred along with the intended style. To address this issue, we 1) extend classifier-free guidance (CFG) to utilize swapping self-attention and propose 2) negative visual query guidance (NVQG) to reduce the transfer of unwanted contents. NVQG employs negative score by intentionally simulating content leakage scenarios that swap queries instead of key and values of self-attention layers from visual style prompts. This simple yet effective method significantly reduces content leakage. Furthermore, we provide careful solutions for using a real image as visual style prompts.Through extensive evaluation across various styles and text prompts, our method demonstrates superiority over existing approaches, reflecting the style of the references, and ensuring that resulting images match the text prompts.

Preference optimization algorithms typically enhance LLM response quality by leveraging human feedback on multiple answers given a fixed instruction. However, these methods often lack capturing the dynamic nature of conversational exchanges. For large vision-language models (LVLMs), direct preference optimization (DPO) can over-emphasize linguistic nuances while overlooking visual context. To address this challenge, we introduce structured policy optimization (SPO) -- a novel preference optimization method that simultaneously aligns preference instructions, responses, and dialogue interactions to improve multi-modal understanding and reasoning capabilities. The efficacy of SPO is attributed to one key design:treating the questioning and answering as a sequential action and binding them through a trajectory reward. This reward formulation better aligns with real-world dialogue studies and eliminates the need for fixed instructions. We evaluate our models on interleaved benchmarks, including image, multi-image, and video-based understanding and reasoning tasks. Experimental results show that the proposed SPO fine-tuning LVLM with multi-modal preference data can align with human preference more efficiently than DPO.

3D medical image self-supervised learning (mSSL) holds great promise for medical analysis. Effectively supporting broader applications requires considering anatomical structure variations in location, scale, and morphology, which are crucial for capturing meaningful distinctions. However, previous mSSL methods partition images with fixed-size patches, often ignoring the structure variations. In this work, we introduce a novel perspective on 3D medical images with the goal of learning structure-aware representations. We assume that patches within the same structure share the same semantics (semantic consistency) while those from different structures exhibit distinct semantics (semantic discrepancy). Based on this assumption, we propose an mSSL framework named S2DC, achieving Structure-aware Semantic Discrepancy and Consistency in two steps. First, S2DC enforces distinct representations for different patches to increase semantic discrepancy by leveraging an optimal transport strategy. Second, S2DC advances semantic consistency at the structural level based on neighborhood similarity distribution. By bridging patch-level and structure-level representations, S2DC achieves structure-aware representations. Thoroughly evaluated across 10 datasets, 4 tasks, and 3 modalities, our proposed method consistently outperforms the state-of-the-art methods in mSSL. The code is available at https://github.com/Ashespt/S2DC.

We present a diffusion-based portrait shadow removal approach that can robustly produce high-fidelity results. Unlike previous methods, we cast shadow removal as diffusion-based inpainting. To this end, we first train a shadow-independent structure extraction network on a real-world portrait dataset with various synthetic lighting conditions, which allows to generate a shadow-independent structure map including facial details while excluding the unwanted shadow boundaries. The structure map is then used as condition to train a structure-guided inpainting diffusion model for removing shadows in a generative manner. Finally, to restore the fine-scale details (e.g., eyelashes, moles and spots) that may not be captured by the structure map, we take the gradients inside the shadow regions as guidance and train a detail restoration diffusion model to refine the shadow removal result. Extensive experiments on the benchmark datasets show that our method clearly outperforms existing methods, and is effective to avoid previously common issues such as facial identity tampering, shadow residual, color distortion, structure blurring, and loss of details. Our code is available at https://github.com/wanchang-yu/Structure-Guided-Diffusion-for-Portrait-Shadow-Removal.

Vision Foundation Models (VFMs) have delivered remarkable performance in Domain Generalized Semantic Segmentation (DGSS). However, recent methods often overlook the fact that visual cues are susceptible, whereas the underlying geometry remains stable, rendering depth information more robust. In this paper, we investigate the potential of integrating depth information with features from VFMs, to improve the geometric consistency within an image and boost the generalization performance of VFMs. We propose a novel fine-tuning DGSS framework, named DepthForge, which integrates the visual cues from frozen DINOv2 or EVA02 and depth cues from frozen Depth Anything V2. In each layer of the VFMs, we incorporate depth-aware learnable tokens to continuously decouple domain-invariant visual and spatial information, thereby enhancing depth awareness and attention of the VFMs. Finally, we develop a depth refinement decoder and integrate it into the model architecture to adaptively refine multi-layer VFM features and depth-aware learnable tokens. Extensive experiments are conducted based on various DGSS settings and five different datasets as unseen target domains. The qualitative and quantitative results demonstrate that our method significantly outperforms alternative approaches with stronger performance, steadier visual-spatial attention, and superior generalization ability. In particular, DepthForge exhibits outstanding performance under extreme conditions (e.g., night and snow). Code is available at https://github.com/SY-Ch/DepthForge.

Generating sketches guided by reference styles requires precise transfer of stroke attributes, such as line thickness, deformation, and texture sparsity, while preserving semantic structure and content fidelity. To this end, we propose Stroke2Sketch, a novel training-free framework that introduces cross-image stroke attention, a mechanism embedded within self-attention layers to establish fine-grained semantic correspondences and enable accurate stroke attribute transfer. This allows our method to adaptively integrate reference stroke characteristics into content images while maintaining structural integrity. Additionally, we develop adaptive contrast enhancement and semantic-focused attention to reinforce content preservation and foreground emphasis. Stroke2Sketch effectively synthesizes stylistically faithful sketches that closely resemble handcrafted results, outperforming existing methods in expressive stroke control and semantic coherence.

Realistic scene reconstruction in driving scenarios poses significant challenges due to fast-moving objects. Most existing methods rely on labor-intensive manual labeling of object poses to reconstruct dynamic objects in canonical space and move them based on these poses during rendering. While some approaches attempt to use 3D object trackers to replace manual annotations, the limited generalization of 3D trackers - caused by the scarcity of large-scale 3D datasets - results in inferior reconstructions in real-world settings. In contrast, 2D foundation models demonstrate strong generalization capabilities. To eliminate the reliance on 3D trackers and enhance robustness across diverse environments, we propose a stable object tracking module by leveraging associations from 2D deep trackers within a 3D object fusion strategy. We address inevitable tracking errors by further introducing a motion learning strategy in an implicit feature space that autonomously corrects trajectory errors and recovers missed detections. Experimental results on Waymo-NOTR and KITTI show that our method outperforms existing approaches. Our code will be made publicly available.

Denoising diffusion models have emerged as powerful tools for image manipulation, yet interactive, localized editing workflows remain underdeveloped. We introduce Layered Diffusion Brushes (LDB), a novel training-free framework that enables interactive, layer-based editing using standard diffusion models. LDB defines each "layer" as a self-contained set of parameters guiding the generative process, enabling independent, non-destructive, and fine-grained prompt-guided edits, even in overlapping regions. LDB leverages a unique intermediate latent caching approach to reduce each edit to only a few denoising steps, achieving 140 ms per edit on consumer GPUs. An editor implementing LDB, incorporating familiar layer concepts, was evaluated via user study and quantitative metrics. Results demonstrate LDB's superior speed alongside comparable or improved image quality, background preservation, and edit fidelity relative to state-of-the-art methods across various sequential image manipulation tasks. The findings highlight LDB's ability to significantly enhance creative workflows by providing an intuitive and efficient approach to diffusion-based image editing and its potential for expansion into related subdomains, such as video editing.

We introduce ProVideLLM, an end-to-end framework for real-time procedural video understanding. ProVideLLM integrates a multimodal cache configured to store two types of tokens -- verbalized text tokens, which provide compressed textual summaries of long-term observations, and visual tokens, encoded with DETR-QFormer to capture fine-grained details from short-term observations. This design reduces token count by 22xover existing methods when representing one hour of long-term observations while effectively encoding fine-granularity of the present. By interleaving these tokens in our multimodal cache, ProVideLLM achieves sub-linear scaling of memory and compute with video length, ensuring per-frame streaming inference at 10 FPS and streaming dialogue at 25 FPS, with a minimal 2GB GPU memory footprint. ProVideLLM also sets new state-of-the-art results on six procedural tasks across four datasets.

With the rise of real-world human-AI interaction applications, such as AI assistants, the need for Streaming Video Dialogue is critical. To address this need, we introduce StreamMind, a video LLM framework that achieves ultra-FPS streaming video processing (100 fps on a single A100) and enables proactive, always-on responses in real time, without explicit user intervention. To solve the key challenge of the contradiction between linear video streaming speed and quadratic transformer computation cost, we propose a novel perception-cognition interleaving paradigm named "event-gated LLM invocation", in contrast to the existing per-time-step LLM invocation. By introducing a Cognition Gate network between the video encoder and the LLM, LLM is only invoked when relevant events occur. To realize the event feature extraction with constant cost, we propose Event-Preserving Feature Extractor (EPFE) based on state-space method, generating a single perception token for spatiotemporal features. These techniques enable the video LLM with full-FPS perception and real-time cognition response. Experiments on Ego4D and SoccerNet streaming tasks, as well as standard offline benchmarks, demonstrate state-of-the-art performance in both model capability and real-time efficiency, paving the way for ultra-high-FPS applications, such as Game AI Copilot and interactive media.

The advent of 3D Gaussian Splatting (3DGS) has advanced 3D scene reconstruction and novel view synthesis. With the growing interest of interactive applications that need immediate feedback, online 3DGS reconstruction in real-time is in high demand. However, none of existing methods yet meet the demand due to three main challenges: the absence of predetermined camera parameters, the need for generalizable 3DGS optimization, and the necessity of reducing redundancy. We propose StreamGS, an online generalizable 3DGS reconstruction method for unposed image streams, which progressively transform image streams to 3D Gaussian streams by predicting and aggregating per-frame Gaussians. Our method overcomes the limitation of the initial point reconstruction in tackling out-of-domain (OOD) issues by introducing a content adaptive refinement. The refinement enhances cross-frame consistency by establishing reliable pixel correspondences between adjacent frames. Such correspondences further aid in merging redundant Gaussians through cross-frame feature aggregation. The density of Gaussians is thereby reduced, empowering online reconstruction by significantly lowering computational and memory costs. Extensive experiments on diverse datasets have demonstrated that StreamGS achieves quality on par with optimization-based approaches but does so 150 times faster, and exhibits superior generalizability in handling OOD scenes.

We introduce StreamDiffusion, a real-time diffusion pipeline designed for streaming image generation. Existing diffusion models are adept at creating images from text or image prompts, yet they often fall short in real-time interaction. This limitation becomes particularly evident in scenarios involving continuous input, such as augmented/virtual reality, video game graphics rendering, live video streaming, and broadcasting, where high throughput is imperative. StreamDiffusion tackles this challenge through a novel pipeline-level system design. It employs unique strategies like batching the denoising process (Stream Batch), residual classifier-free guidance (R-CFG), and stochastic similarity filtering (SSF). Additionally, it seamlessly integrates advanced acceleration technologies for maximum efficiency. Specifically, Stream Batch reformulates the denoising process by eliminating the traditional wait-and-execute approach and utilizing a batching denoising approach, facilitating fluid and high-throughput streams. This results in 1.5x higher throughput compared to the conventional sequential denoising approach. R-CFG significantly addresses inefficiencies caused by repetitive computations during denoising. It optimizes the process to require minimal or no additional computations, leading to speed improvements of up to 2.05x compared to previous classifier-free methods. Besides, our stochastic similarity filtering dramatically lowers GPU activation frequency by halting computations for static image flows, achieving a remarkable reduction in computational consumption--2.39 times on an RTX 3060 GPU and 1.99 times on an RTX 4090 GPU, respectively. The synergy of our proposed strategies with established acceleration technologies enables image generation to reach speeds of up to 91.07 fps on a single RTX 4090 GPU, outperforming the throughput of AutoPipeline, developed by Diffusers, by more than 59.56x.

While haircut indicates distinct personality, existing avatar generation methods fail to model practical hair due to the data limitation or entangled representation. We propose StrandHead, a novel text-driven method capable of generating 3D hair strands and disentangled head avatars with strand-level attributes. Instead of using large-scale hair-text paired data for supervision, we demonstrate that realistic hair strands can be generated from prompts by distilling 2D generative models pre-trained on human mesh data. To this end, we propose a meshing approach guided by strand geometry to guarantee the gradient flow from the distillation objective to the neural strand representation. The optimization is then regularized by statistically significant haircut features, leading to stable updating of strands against unreasonable drifting. These employed 2D/3D human-centric priors contribute to text-aligned and realistic 3D strand generation. Extensive experiments show that StrandHead achieves the state-of-the-art performance on text to strand generation and disentangled 3D head avatar modeling. The generated 3D hair can be applied on avatars for strand-level editing, as well as implemented in the graphics engine for physical simulation or other applications. Project page: https://xiaokunsun.github.io/StrandHead.github.io/.