Explicitly disentangling style and content in vision models remains challenging due to their semantic overlap and the subjectivity of human perception. Existing methods propose separation through generative or discriminative objectives, but they still face the inherent ambiguity of disentangling intertwined concepts. Instead, we ask: Can we bypass explicit disentanglement by learning to merge style and content invertibly, allowing separation to emerge naturally? We propose SCFlow, a flow-matching framework that learns bidirectional mappings between entangled and disentangled representations. Our approach is built upon three key insights: 1) Training solely to merge style and content, a well-defined task, enables invertible disentanglement without explicit supervision; 2) flow matching bridges on arbitrary distributions, avoiding the restrictive Gaussian priors of diffusion models and normalizing flows; and 3) a synthetic dataset of 510,000 samples was curated to simulate disentanglement through systematic style-content pairing. Beyond controllable generation tasks, we demonstrate that SCFlow generalizes to ImageNet-1k and WikiArt in zero-shot settings and achieves competitive performance, highlighting that disentanglement naturally emerges from the invertible merging process. Code and dataset: https://github.com/CompVis/SCFlow
论文检索
输入标题、作者或关键词，从 2,701 篇学术成果中精准定位
Multimodal large language models (MLLMs) are increasingly used to evaluate text-to-image (TTI) generation systems, providing automated judgments based on visual and textual context. However, these "judge" models often suffer from biases, overconfidence, and inconsistent performance across diverse image domains. While prompt ensembling has shown promise for mitigating these issues in unimodal, text-only settings, our experiments reveal that standard ensembling methods fail to generalize effectively for TTI tasks. To address these limitations, we propose a new multimodal-aware method called **M**ultimodal **M**ixture-of-**B**ayesian Prompt Ensembles (MMB). Our approach uses a Bayesian prompt ensemble approach augmented by image clustering, allowing the judge to dynamically assign prompt weights based on the visual characteristics of each sample. We show that MMB improves accuracy in pairwise preference judgments and greatly enhances calibration, making it easier to gauge the judge's true uncertainty. In evaluations on two TTI benchmarks, HPSv2 and MJBench, MMB outperforms existing baselines in alignment with human annotations and calibration across varied image content. Our findings highlight the importance of multimodal-specific strategies for judge calibration and suggest a promising path forward for reliable large-scale TTI evaluation.
Mitigating biases in computer vision models is an essential step towards trustworthy artificial intelligence systems. Existing bias mitigation methods are limited to predefined biases, preventing their use in visual datasets where multiple, possibly unknown biases exist. To address this limitation, we introduce MAVias, an open-set bias mitigation approach that leverages foundation models to discover spurious associations between visual attributes and target classes. MAVias first captures a wide variety of visual features in natural language via a foundation image tagging model, and then leverages a large language model to select visual features that define the target class, resulting in a set of language-coded potential visual biases. It then translates these biases into vision-language embeddings and introduces an in-processing bias mitigation approach to prevent the model from encoding information related to them. Experiments on diverse datasets, including CelebA, Waterbirds, ImageNet, and UrbanCars, show that MAVias effectively detects and mitigates a wide range of biases in visual recognition tasks, outperforming current state-of-the-art.
3D occupancy prediction provides a comprehensive description of the surrounding scenes and has become an essential task for 3D perception. Most existing methods focus on offline perception from one or a few views and cannot be applied to embodied agents that demand to gradually perceive the scene through progressive embodied exploration. In this paper, we formulate an embodied 3D occupancy prediction task to target this practical scenario and propose a Gaussian-based EmbodiedOcc framework to accomplish it. We initialize the global scene with uniform 3D semantic Gaussians and progressively update local regions observed by the embodied agent. For each update, we extract semantic and structural features from the observed image and efficiently incorporate them via deformable cross-attention to refine the regional Gaussians. Finally, we employ Gaussian-to-voxel splatting to obtain the global 3D occupancy from the updated 3D Gaussians. Our EmbodiedOcc assumes an unknown (i.e., uniformly distributed) environment and maintains an explicit global memory of it with 3D Gaussians. It gradually gains knowledge through the local refinement of regional Gaussians, which is consistent with how humans understand new scenes through embodied exploration. We reorganize an EmbodiedOcc-ScanNet benchmark based on local annotations to facilitate the evaluation of the embodied 3D occupancy prediction task. Our EmbodiedOcc outperforms existing methods by a large margin and accomplishes the embodied occupancy prediction with high accuracy and efficiency. Code: https://github.com/YkiWu/EmbodiedOcc.
Towards Robust Defense against Customization via Protective Perturbation Resistant to Diffusion-based Purification
PDF ↗Diffusion models like Stable Diffusion have become prominent in visual synthesis tasks due to their powerful customization capabilities, which also introduce significant security risks, including deepfakes and copyright infringement. In response, a class of methods known as protective perturbation emerged, which mitigates image misuse by injecting imperceptible adversarial noise. However, purification can remove protective perturbations, thereby exposing images again to the risk of malicious forgery. In this work, we formalize the anti-purification task, highlighting challenges that hinder existing approaches, and propose a simple diagnostic protective perturbation named AntiPure. AntiPure exposes vulnerabilities of purification within the "purification-customization" workflow, owing to two guidance mechanisms: 1) Patch-wise Frequency Guidance, which reduces the model's influence over high-frequency components in the purified image, and 2) Erroneous Timestep Guidance, which disrupts the model's denoising strategy across different timesteps. With additional guidance, AntiPure embeds imperceptible perturbations that persist under representative purification settings, achieving effective post-customization distortion. Experiments show that, as a stress test for purification, AntiPure achieves minimal perceptual discrepancy and maximal distortion, outperforming other protective perturbation methods within the purification-customization workflow.
Current state-of-the-art methods for skeleton-based temporal action segmentation are predominantly supervised and require annotated data, which is expensive to collect. In contrast, existing unsupervised temporal action segmentation methods have focused primarily on video data, while skeleton sequences remain underexplored, despite their relevance to real-world applications, robustness, and privacy-preserving nature. In this paper, we propose a novel approach for unsupervised skeleton-based temporal action segmentation. Our method utilizes a sequence-to-sequence temporal autoencoder that keeps the information of the different joints disentangled in the embedding space. Latent skeleton sequences are then divided into non-overlapping patches and quantized to obtain distinctive skeleton motion words, driving the discovery of semantically meaningful action clusters. We thoroughly evaluate the proposed approach on three widely used skeleton-based datasets, namely HuGaDB, LARa, and BABEL. The results demonstrate that our model outperforms the current state-of-the-art unsupervised temporal action segmentation methods.
We introduce SPFSplat, an efficient framework for 3D Gaussian splatting from sparse multi-view images, requiring no ground-truth poses during training or inference. It employs a shared feature extraction backbone, enabling simultaneous prediction of 3D Gaussian primitives and camera poses in a canonical space from unposed inputs within a single feed-forward step. Alongside the rendering loss based on estimated novel-view poses, a reprojection loss is integrated to enforce the learning of pixel-aligned Gaussian primitives for enhanced geometric constraints. This pose-free training paradigm and efficient one-step feed-forward design make SPFSplat well-suited for practical applications. Remarkably, despite the absence of pose supervision, SPFSplat achieves state-of-the-art performance in novel view synthesis even under significant viewpoint changes and limited image overlap. It also surpasses recent methods trained with geometry priors in relative pose estimation. Code and trained models are available on our project page: https://ranrhuang.github.io/spfsplat/.
In this paper, we introduce ILLUME, a unified multimodal large language model (MLLM) that seamlessly integrates multimodal understanding and generation capabilities within a single large language model through a unified next-token prediction formulation.To address the large dataset size typically required for image-text alignment, we propose to enhance data efficiency through the design of a vision tokenizer that incorporates semantic information and a progressive multi-stage training procedure. This approach reduces the dataset size to just 15M for pretraining -- over four times fewer than what is typically needed -- while achieving competitive or even superior performance with existing unified MLLMs, such as Janus. Additionally, to promote synergistic enhancement between understanding and generation capabilities, which is under-explored in previous works, we introduce a novel self-enhancing multimodal alignment scheme. This scheme supervises the MLLM to self-assess the consistency between text descriptions and self-generated images, facilitating the model to interpret images more accurately and avoid unrealistic and incorrect predictions caused by misalignment in image generation. Based on our extensive experiments, our proposed ILLUME stands out and competes with state-of-the-art unified MLLMs and specialized models across various benchmarks for multimodal understanding, generation, and editing.
Reflective flares are common artifacts in photography that degrade image quality, introducing in-focus flares, which appear as bright, regular spot patterns, and out-of-focus flares, which are diffuse and semi-transparent, obscuring the underlying scene. While previous methods have achieved some success in removing in-focus flares, they struggle with the diffuse nature of out-of-focus flares. The lack of an out-of-focus flare dataset has further hindered the development of effective flare removal models. In this work, we construct a large-scale out-of-focus flare dataset generated with Blender. We propose to use a color alignment approach on diffusion models to address the challenges of out-of-focus reflective flare removal. Rather than fully reconstructing flare-affected regions, our method adjusts the color distribution to reduce artifact visibility while preserving image content. Specifically, we apply a differentiable histogram loss on the diffusion model, which is derived from the Earth Mover's Distance (EMD), to effectively align color distributions. The proposed approach, trained exclusively on synthetic data, outperforms existing methods on both synthetic and real-world data, demonstrating improved performance in flare removal.
TCFG: Truncated Classifier-Free Guidance for Efficient and Scalable Text-to-Image Acceleration
PDF ↗Diffusion models have achieved remarkable success in image and video generation due to their powerful generative capabilities. However, they suffer from slow inference speed and high computational costs. Existing acceleration methods for diffusion models may compromise model performance and struggle to generalize across diverse diffusion model architectures and downstream tasks. To address these issues, we propose a model-agnostic and highly scalable acceleration strategy for text-controlled image generation. Specifically, we dynamically modulate the text guidance coefficience and truncate redundant text-related computations during the denoising process. Experimental results demonstrate that our approach achieves significant model acceleration while preserving precise text-image alignment, showcasing the potential for a wide range of diffusion models and downstream applications.
Recent advances in video generation can produce realistic, minute-long single-shot videos with scalable diffusion transformers. However, real-world narrative videos require multi-shot scenes with visual and dynamic consistency across shots. In this work, we introduce Long Context Tuning (LCT), a training paradigm that expands the context window of pre-trained single-shot video diffusion models to learn scene-level consistency directly from data. Our method expands full attention mechanisms from individual shots to encompass all shots within a scene, incorporating interleaved 3D position embedding and an asynchronous noise strategy, enabling both joint and auto-regressive shot generation without additional parameters. Models with bidirectional attention after LCT can further be fine-tuned with context-causal attention, facilitating auto-regressive generation with efficient KV-cache. Experiments demonstrate single-shot models after LCT can produce coherent multi-shot scenes and exhibit emerging capabilities, including compositional generation and interactive shot extension, paving the way for more practical visual content creation.
SV4D 2.0: Enhancing Spatio-Temporal Consistency in Multi-View Video Diffusion for High-Quality 4D Generation
PDF ↗We present Stable Video 4D 2.0 (SV4D 2.0), a multi-view video diffusion model for dynamic 3D asset generation. Compared to its predecessor SV4D, SV4D 2.0 is more robust to occlusions and large motion, generalizes better to real-world videos, and produces higher-quality outputs in terms of detail sharpness and spatio-temporal consistency. We achieve this by introducing key improvements in multiple aspects: 1) network architecture: eliminating the dependency of reference multi-views and designing blending mechanism for 3D and frame attention, 2) data: enhancing quality and quantity of training data, 3) training strategy: adopting progressive 3D-4D training for better generalization, and 4) 4D optimization: handling 3D inconsistency and large motion via 2-stage refinement and progressive frame sampling. Extensive experiments demonstrate significant performance gain by SV4D 2.0 both visually and quantitatively, achieving better detail (-14% LPIPS) and 4D consistency (-44% FV4D) in novel-view video synthesis and 4D optimization (-12% LPIPS and -24% FV4D) compared to SV4D.
The Best of Both Worlds: Integrating Language Models and Diffusion Models for Video Generation
PDF ↗Recent advancements in text-to-video (T2V) generation have been driven by two competing paradigms: autoregressive language models and diffusion models. However, each paradigm has intrinsic limitations: language models struggle with visual quality and error accumulation, while diffusion models lack semantic understanding and causal modeling. In this work, we propose LanDiff, a hybrid framework that synergizes the strengths of both paradigms through coarse-to-fine generation. Our architecture introduces three key innovations: (1) a semantic tokenizer that compresses 3D visual features into compact 1D discrete representations through efficient semantic compression, achieving a ~14,000xcompression ratio; (2) a language model that generates semantic tokens with high-level semantic relationships; (3) a streaming diffusion model that refines coarse semantics into high-fidelity videos. Experiments show that LanDiff, a 5B model, achieves a score of 85.43 on the VBench T2V benchmark, surpassing the state-of-the-art open-source models Hunyuan Video (13B) and other commercial models such as Sora, Kling, and Hailuo. Furthermore, our model also achieves state-of-the-art performance in long video generation, surpassing other open-source models in this field. Our demo can be viewed at https://landiff.github.io/.
Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA have transformed vision model adaptation, enabling the rapid deployment of customized models. However, the compactness of LoRA adaptations introduces new safety concerns, particularly their vulnerability to model extraction attacks. This paper introduces a new focus of model extraction attacks named LoRA extraction that extracts LoRA-adaptive models based on a public pre-trained model. We then propose a novel extraction method called StolenLoRA which trains a substitute model to extract the functionality of a LoRA-adapted model using synthetic data. StolenLoRA leverages a Large Language Model to craft effective prompts for data generation, and it incorporates a Disagreement-based Semi-supervised Learning (DSL) strategy to maximize information gain from limited queries.Our experiments demonstrate the effectiveness of StolenLoRA, achieving up to a 96.60% attack success rate with only 10k queries, even in cross-backbone scenarios where the attacker and victim models utilize different pre-trained backbones. These findings reveal the specific vulnerability of LoRA-adapted models to this type of extraction and underscore the urgent need for robust defense mechanisms tailored to PEFT methods.We also explore a preliminary defense strategy based on diversified LoRA deployments, highlighting its potential to mitigate such attacks.
Positron Emission Tomography (PET) is a powerful molecular imaging tool that plays a crucial role in modern medical diagnostics by visualizing radio-tracer distribution to reveal physiological processes. Accurate organ segmentation from PET images is essential for comprehensive multi-systemic analysis of interactions between different organs and pathologies.Existing segmentation methods are limited by insufficient annotation data and varying levels of annotation, resulting in weak generalization ability and difficulty in clinical application. Recent developments in segmentation foundation models have shown superior versatility across diverse segmentation tasks. Despite the efforts of medical adaptations, these works primarily focus on structural medical images with detailed physiological structural information and exhibit limited generalization performance on molecular PET imaging. In this paper, we collect and construct PETS-5k, the largest PET segmentation dataset to date, comprising 5,731 three-dimensional whole-body PET images and encompassing over 1.3M 2D images. Based on the established dataset, we develop SegAnyPET, a modality-specific 3D foundation model for universal promptable segmentation from PET images. To issue the challenge of discrepant annotation quality, we adopt a cross prompting confident learning (CPCL) strategy with an uncertainty-guided self-rectification process to robustly learn segmentation from high-quality labeled data and low-quality noisy labeled data for promptable segmentation. Experimental results demonstrate that SegAnyPET can segment seen and unseen target organs using only one or a few prompt points, outperforming state-of-the-art foundation models and task-specific fully supervised models with higher accuracy and strong generalization ability for universal segmentation. As the first foundation model for PET images, we believe that SegAnyPET will advance the applications to various downstream tasks for molecular imaging.
Graphic design is crucial for conveying ideas and messages. Designers usually organize their work into objects, backgrounds, and vectorized text layers to simplify editing. However, this workflow demands considerable expertise. With the rise of GenAI methods, an endless supply of high-quality graphic designs in pixel format has become more accessible, though these designs often lack editability. Despite this, non-layered designs still inspire human designers, influencing their choices in layouts and text styles, ultimately guiding the creation of layered designs. Motivated by this observation, we propose Accordion, a graphic design generation framework taking the first attempt to convert AI-generated designs into editable layered designs, meanwhile refining nonsensical AI-generated text with meaningful alternatives guided by user prompts. It is built around a vision language model (VLM) playing distinct roles in three curated stages: (1) reference creation, (2) design planning, and (3) layer generation. For each stage, we design prompts to guide the VLM in executing different tasks. Distinct from existing bottom-up methods (e.g., COLE and Open-COLE) that gradually generate elements to create layered designs, our approach works in a top-down manner by using the visually harmonious reference image as global guidance to decompose each layer. Additionally, it leverages multiple vision experts such as SAM and element removal models to facilitate the creation of graphic layers. We train our method using the in-house graphic design dataset Design39K, augmented with AI-generated design images coupled with refined ground truth created by a customized inpainting model. Experimental results and user studies by designers show that Accordion generates favorable results on the DesignIntention benchmark, including tasks such as text-to-template, adding text to background, and text de-rendering, and also excels in creating design variations.
Multi-image Interleaved Reasoning aims to improve Multimodal Large Language Models' (MLLMs) ability to jointly comprehend and reason across multiple images and their associated textual contexts, introducing unique challenges beyond single-image or non-interleaved multi-image tasks.While current multi-image benchmarks overlook interleaved textual contexts and neglect distinct relationships between individual images and their associated texts, enabling models to reason over multi-image interleaved data may significantly enhance their comprehension of complex scenes and better capture cross-modal correlations.To bridge this gap, we introduce a novel benchmark MIR, requiring joint reasoning over multiple images accompanied by interleaved textual contexts to accurately associate image regions with corresponding texts and logically connect information across images.To enhance MLLMs' ability to comprehend multi-image interleaved data, we introduce reasoning steps for each instance within the benchmark and propose a stage-wise curriculum learning strategy. This strategy follows an "easy to hard" approach, progressively guiding models from simple to complex scenarios, thereby enhancing their ability to handle challenging tasks.Extensive experiments benchmarking multiple MLLMs demonstrate that our method significantly enhances models' reasoning performance on MIR and other established benchmarks, highlighting the challenges current MLLMs face with multi-image interleaved reasoning.We believe that MIR will encourage further research into multi-image interleaved reasoning, facilitating advancements in MLLMs' capability to handle complex inter-modal tasks.
Model performance in text-to-image (T2I) and image-to-image (I2I) generation often depends on multiple aspects, including quality, alignment, diversity, and robustness. However, models' complex trade-offs among these dimensions have been rarely explored due to (1) the lack of datasets that allow fine-grained quantification of these trade-offs, and (2) using a single metric for multiple dimensions. To address this gap, we introduce TRIG-Bench (Trade-offs in Image Generation), which spans 10 dimensions (Realism, Originality, Aesthetics, Content, Relation, Style, Knowledge, Ambiguity, Toxicity and Bias), contains over 40,200 samples, and covers 132 Pairwise Dimensional Subsets. Furthermore, we develop TRIGScore,a VLM-as-judge metric that automatically adapts to various dimensions. Based on this, we evaluate 14 cutting-edge models across T2I and I2I tasks. In addition, we propose the Relation Recognition System and generate the Dimension Trade-off Map (DTM), which visualizes model-specific capability trade-offs. Our experiments demonstrate that DTM consistently provides a comprehensive understanding of the trade-offs between dimensions for each type of generation models. Notably, after fine-tuning on DTM, the model's dimension-specific impact is mitigated, and overall performance is enhanced. Code is available at: https://github.com/fesvhtr/TRIG.
Autoregressive visual generation models typically rely on tokenizers to compress images into tokens that can be predicted sequentially. A fundamental dilemma exists in token representation: discrete tokens enable straightforward modeling with standard cross-entropy loss, but suffer from information loss and tokenizer training instability; continuous tokens better preserve visual details, but require complex distribution modeling, complicating the generation pipeline. In this paper, we propose TokenBridge, which bridges this gap by maintaining the strong representation capacity of continuous tokens while preserving the modeling simplicity of discrete tokens. To achieve this, we decouple discretization from the tokenizer training process through post-training quantization that directly obtains discrete tokens from continuous representations. Specifically, we introduce a dimension-wise quantization strategy that independently discretizes each feature dimension, paired with a lightweight autoregressive prediction mechanism that efficiently model the resulting large token space. Extensive experiments show that our approach achieves reconstruction and generation quality on par with continuous methods while using standard categorical prediction. This work demonstrates that bridging discrete and continuous paradigms can effectively harness the strengths of both approaches, providing a promising direction for high-quality visual generation with simple autoregressive modeling. Project page: https://yuqingwang1029.github.io/TokenBridge
ReCoT: Reflective Self-Correction Training for Mitigating Confirmation Bias in Large Vision-Language Models
PDF ↗Recent advancements in Large Vision-Language Models (LVLMs) have greatly improved their ability to understand both visual and text information. However, a common problem in LVLMs is confirmation bias, where models tend to repeat previous assumptions and follow earlier viewpoints instead of reflecting and correcting themselves. This problem is more common in smaller-scale LVLMs, as they are usually fine-tuned with training data that is mostly positive, focusing on generating coherent dialogue. To address this issue, we introduce ReCoT, a method designed to mitigate confirmation bias in smaller-scale LVLMs through Reflective Self-Correction Training.The method follows a two-stage SFT-DPO paradigm: the first SFT stage aims to cultivate the model's reflective correction abilities, while the DPO stage focuses on enhancing the consistency between answers and reflections. Specifically, we construct dialogue-based reflective samples, which serve as adversarial samples during SFT. In this process, the model is initially presented with a potentially incorrect answer, followed by a reflection and correction phase to generate the final answer. To enhance answer-reflection consistency, we propose the consistency direct preference optimization. To comprehensively evaluate the effectiveness of our ReCoT, we introduce a set of novel metrics to measure the accuracy of the reflection and correction process. Extensive experiments show that ReCoT enables LVLM to engage in robust self-reflection and error correction and reduce confirmation bias.