Text-to-image (T2I) diffusion models such as SDXL and FLUX have achieved impressive photorealism, yet small-scale distortions remain pervasive in limbs, face, text and so on. Existing refinement approaches either perform costly iterative re-generation or rely on vision-language models (VLMs) with weak spatial grounding, leading to semantic drift and unreliable local edits. To close this gap, we propose **Agentic Retoucher**, a hierarchical decision-driven framework that reformulates post-generation correction as a human-like *perception-reasoning-action* loop.Specifically, we design (1) a **perception agent** that learns contextual saliency for fine-grained distortion localization under text-image consistency cues, (2) a **reasoning agent** that performs human-aligned inferential diagnosis via progressive preference alignment, and (3) an **action agent** that adaptively plans localized inpainting guided by user preference. This design integrates perceptual evidence, linguistic reasoning, and controllable correction into a unified, self-corrective decision process. To enable fine-grained supervision and quantitative evaluation, we further construct **GenBlemish-27K**, a dataset of 6K T2I images with 27K annotated artifact regions across 12 categories.Extensive experiments demonstrate that Agentic Retoucher consistently outperforms state-of-the-art methods in perceptual quality, distortion localization and human preference alignment, establishing a new paradigm for self-corrective and perceptually reliable T2I generation.
论文检索
输入标题、作者或关键词，从 9,654 篇学术成果中精准定位
Self-consistency has proven to be an effective technique for improving LLM performance on natural language reasoning tasks in a lightweight, unsupervised manner. In this work, we study how to adapt self-consistency to visual domains. Specifically, we consider the generation and verification of LLM-produced motion graphics trajectories. Given a prompt (e.g., "Move the circle in a spiral path"), we first sample diverse motion trajectories from an LLM, and then identify groups of consistent trajectories via clustering. Our key insight is to model the family of shapes associated with a prompt as a prototype trajectory paired with a group of geometric transformations (e.g., rigid, similarity, and affine). Two trajectories can then be considered consistent if one can be transformed into the other under the warps allowable by the transformation group. We propose an algorithm that automatically recovers a shape family, using hierarchical relationships between a set of candidate transformation groups. Our approach improves the accuracy of LLM-based trajectory generation by 4-6%. We further extend our method to support verification, observing 11% precision gains over VLM baselines. Our code and dataset are available at https://majiaju.io/trajectory-self-consistency
Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMs
PDF ↗Although reinforcement learning (RL) has significantly advanced reasoning capabilities in large multimodal language models (MLLMs), its efficacy remains limited for lightweight models essential for edge deployments. To address this issue, we leverage causal analysis and experiment to reveal the underlying phenomenon of perceptual bias, demonstrating that RL-based fine-tuning compels lightweight models to preferentially adopt perceptual shortcuts induced by data biases, rather than developing genuine reasoning abilities. Motivated by this insight, we propose VideoThinker, a causal-inspired framework that cultivates robust reasoning in lightweight models through a two-stage debiasing process. First, the Bias Aware Training stage forges a dedicated "bias model" to embody these shortcut behaviors. Then, the Causal Debiasing Policy Optimization (CDPO) algorithm fine-tunes the primary model, employing an innovative repulsive objective to actively push it away from the bias model's flawed logic while simultaneously pulling it toward correct, generalizable solutions. Our model, VideoThinker-R1, establishes a new state-of-the-art in video reasoning efficiency. For same-scale comparison, requiring no Supervised Fine-Tuning (SFT) and using only 1% of the training data for RL, it surpasses VideoRFT-3B with a 3.2% average gain on widely-used benchmarks and a 7% lead on VideoMME. For cross-scale comparison, it outperforms the larger Video-UTR-7B model on multiple benchmarks, including a 2.1% gain on MVBench and a 3.8% gain on TempCompass. Code is available at https://github.com/falonss703/VideoThinker.
Federated Graph Learning (FGL) offers a privacy-preserving paradigm for collaborative training on graph data, yet significant topological heterogeneity poses a critical threat to generalization fairness, often yielding a global model dominated by a subset of clients. This introduces two critical issues: at the global level, aggregation bias disproportionately amplifies the influence of dominant clients, while at the local level, blind optimization results in inefficient and inequitable training processes. To address these challenges, we propose FedSST, an adaptive fairness framework. FedSST introduces a fair, structure-based signal to quantify client contributions, which in turn guides fair aggregation and adaptive local training. Extensive experiments across diverse cross-domain and cross-dataset settings demonstrate that FedSST enhances generalization fairness and overall model performance, outperforming various state-of-the-art methods.
MapReduce LoRA: Advancing the Pareto Front in Multi-Preference Optimization for Generative Models
PDF ↗Reinforcement learning from human feedback (RLHF) with reward models has advanced alignment of generative models to human aesthetic and perceptual preferences. However, jointly optimizing multiple rewards often incurs an alignment tax--improving one dimension while degrading others. To address this, we introduce two complementary methods: MapReduce LoRA and Reward-aware Token Embedding (RaTE). MapReduce LoRA trains preference-specific LoRA experts in parallel and iteratively merges them to refine a shared base model; RaTE learns reward-specific token embeddings that compose at inference for flexible preference control. Experiments on Text-to-Image generation (Stable Diffusion 3.5 Medium and FLUX.1-dev) show improvements of 36.1%, 4.6%, and 55.7%, and 32.7%, 4.3%, and 67.1% on GenEval, PickScore, and OCR, respectively. On Text-to-Video generation (HunyuanVideo), visual and motion quality improve by 48.1% and 90.0%, respectively. On the language task, Helpful Assistant, with Llama-2 7B, helpful and harmless improve by 43.4% and 136.7%, respectively. Our framework sets a new state-of-the-art multi-preference alignment recipe across modalities.
TriSim: Tri-Dimensional Similarity Modeling with Extreme Value Theory for False-Negative Mitigation in Remote Sensing Image-Text Retrieval
PDF ↗In remote sensing (RS) cross-modal retrieval, most existing methods employ contrastive learning as their primary optimization objective, aligning anchors with positive counterparts and distinguishing them from negative samples. To improve negative sampling, these approaches typically set thresholds on cross-modal similarity scores, designating negatives that exceed the threshold as false negative samples (FNS). However, dependence on a single cross-modal similarity threshold is fragile because it fails to account for the cross-modal semantic overlaps and gaps. To address these challenges, we introduce TriSim, a novel image-text retrieval framework that constructs a tri-dimensional negative similarity space <img-img, img-txt, txt-txt> to mitigate the influence of FNS issue. Specifically, considering that FNS appear as anomalies in this space, Extreme Value Theory (EVT) is applied to model the statistical behavior of the tail distribution for FNS selection. Two complementary tail selection strategies are developed: one identifies samples distant from the dense ellipsoidal center, and the other targets upper-right high-similarity extremes. The selected tail samples are regarded as FNS and modeled using a generalized Pareto distribution, with probabilistic weights assigned in the triplet loss. To further refine the selected FNS, intra-modal saliency differences are computed to generate masks that guide the learning of a gain matrix, which amplifies highly discriminative regions and suppresses ambiguous ones. Extensive experiments on two benchmarks demonstrate the superiority of the proposed TriSim framework in mitigating the influence of false negatives in RS image-text retrieval.
Vision-language models (VLMs) have achieved remarkable success across numerous domains, yet they lag significantly in animal behavior understanding due to severe data scarcity. Annotated animal behavior videos are prohibitively expensive and time-consuming to collect, requiring domain expertise and controlled observation conditions. To address this challenge, we leverage structured domain knowledge as an inductive bias from the Neuro Behavior Ontology (NBO), which provides professional annotations, hierarchical behavior structures, and comprehensive semantic coverage. We construct AnimalBand, an NBO-consistent dataset integrating 74,671 videos across multiple species and behaviors with semantic standardization and extended knowledge. Based on this resource, we present EthoCLIP, an ontology-enhanced vision-language contrastive learning framework that embeds ontology semantics through an ontology-aware graph module to capture hierarchical relationships among behaviors and learn structured semantic dependencies. Incorporating ontological information reduces reliance on purely data-driven learning, thereby alleviating needs for large-scale datasets. Extensive experiments validate both our dataset and method. Results demonstrate that EthoCLIP pretrained on AnimalBand substantially improves behavior recognition accuracy and transfer learning performance across diverse benchmarks, confirming that ontology-driven semantic enrichment effectively mitigates data scarcity in animal behavior understanding. Our data and code will be released at https://github.com/PRIS-CV/AnimalBand.
DarkShake-DVS: Event-based Human Action Recognition under Low-light and Shaking Camera Conditions
PDF ↗Human Action Recognition (HAR) is a fundamental computer vision task with diverse real-world applications. Practical deployments often involve low-light environments and unconstrained 6-DoF camera motion, conditions that degrade visual quality, disrupt temporal coherence, and compromise reliability of existing methods. Event cameras, with high low-light sensitivity and microsecond-level temporal resolution, paired with an inertial measurement unit (IMU), present a promising solution. However, current research faces two key challenges: absence of a benchmark integrating low-light conditions, 6-DoF motion, and synchronized IMU data; and lack of effective motion compensation techniques. To address these, we propose Event-IMU Stabilized HAR (EIS-HAR), with two modules. The first is an EIS module that reduces motion blur via a non-linear warping function to reconstruct a motion-compensated input. The second is a HAR module with a four-stage hybrid architecture to efficiently extract spatiotemporal features for accurate action recognition. To alleviate data scarcity, we introduce DarkShake-DVS, the first large-scale event-based HAR benchmark that includes 18,041 real-world clips captured in low light and intense 6-DoF motion, supplemented by synchronized IMU data. Extensive experiments on three datasets demonstrate consistent superiority of EIS-HAR over state-of-the-art methods.
In multimodal large language models (MLLMs), the surge of visual tokens significantly increases the inference time and computational overhead, making them impractical for real-time or resource-constrained applications.Visual token pruning is a promising strategy for reducing the cost of MLLM inference by removing redundant visual tokens.Existing researches usually assume that all attention heads contribute equally to the visual interpretation.However, our study reveals that different heads may capture distinct visual semantics and inherently play distinct roles in visual processing.In light of this observation, we propose HAWK, a head importance-aware visual token pruning method that perceives the varying importance of attention heads in visual tasks to maximize the retention of crucial tokens.By leveraging head importance weights and text-guided attention to assess visual token significance, HAWK effectively retains task-relevant visual tokens while removing redundant ones.The proposed HAWK is entirely training-free and can be seamlessly applied to various MLLMs.Extensive experiments on multiple mainstream vision-language benchmarks demonstrate that HAWK achieves state-of-the-art accuracy. When applied to Qwen2.5-VL, HAWK retains 96.0% of the original accuracy after pruning 80.2% of the visual tokens. Additionally, it reduces end-to-end latency to 74.4% of the original and further decreases GPU memory usage across the tested models.
Concept erasure in text-to-image diffusion models seeks to remove undesired concepts while preserving overall generative capability. Localized erasure methods aim to restrict edits to the spatial region occupied by the target concept. However, we observe that suppressing a concept can unintentionally weaken semantically related neighbor concepts, reducing fidelity in fine-grained domains. We propose Neighbor-Aware Localized Concept Erasure (NLCE), a training-free framework designed to better preserve neighboring concepts while removing target concepts. It operates in three stages: (1) a spectrally-weighted embedding modulation that attenuates target concept directions while stabilizing neighbor concept representations, (2) an attention-guided spatial gate that identifies regions exhibiting residual concept activation, and (3) a spatially-gated hard erasure that eliminates remaining traces only where necessary. This neighbor-aware pipeline enables localized concept removal while maintaining the surrounding concept neighborhood structure. Experiments on fine-grained datasets (Oxford Flowers, Stanford Dogs) show that our method effectively removes target concepts while better preserving closely related categories. Additional results on celebrity identity, explicit content and artistic style demonstrate robustness and generalization to broader erasure scenarios.
Despite recent advances, Vision Language Models (VLMs) still struggle to grasp the dynamics of the world. We note that the ability to reason about a 4D scene, challenging in itself, is further complicated by two factors. First, VLMs observe motion indirectly via its projection onto 2D images. Second, existing datasets fail to disentangle object and camera motion. To address these challenges, we present a QA generation pipeline that focuses on motion-related scene understanding. We take particular care of the entanglement of camera and object motion by casting tracking in both the traditional way and in a novel, fixed reference system, dubbed True-Motion Tracking, which provides an intuitive description of motion. From this pipeline, we generate a large-scale training dataset of 400K samples, 4DP-QA (4D Perception QA), and a 2.2K-sample benchmark, 4DP-QA-Bench. Training existing models on our dataset yields performance improvements on an external benchmark, validating the effectiveness of our method.
Document understanding is a long-standing practical task. Vision-Language Models (VLMs) have gradually become a primary approach in this domain, demonstrating effective performance on single-page tasks. However, their effectiveness diminishes when handling long documents. In such scenarios, clues are often scattered across multiple pages and modalities, and redundancy from lengthy inputs can impair the model's judgment. While retrieval-augmented generation mitigates this issue by filtering for question-relevant content, the retrieved results still contain substantial redundancy. To address these limitations, we propose SLEUTH, a multi-agent framework. Concretely, SLEUTH orchestrates a retriever and four collaborative agents in a coarse-to-fine process. The framework identifies key textual and visual clues within the retrieved pages, filters for salient visual evidence such as tables and charts, and analyzes the query to devise a reasoning strategy. It ultimately synthesizes a distilled, evidence-dense multimodal context to generate the final prediction. SLEUTH is model-agnostic and scalable. When paired with advanced VLM backbones, it consistently improves performance on multiple long-document benchmarks, achieving SOTA results. Ablation studies verify each module's effectiveness and confirm the benefits of our hierarchical refinement paradigm.
As social media platforms proliferate, users increasingly demand intuitive ways to create diverse, high-quality portrait collections. In this work, we introduce Portrait Collection Generation (PCG), a novel task that generates coherent portrait collections by editing a reference portrait image through natural language instructions. This task poses two unique challenges to existing methods: (1) complex multi-attribute modifications such as pose, spatial layout, and camera viewpoint; and (2) high-fidelity detail preservation including identity, clothing, and accessories. To address these challenges, we propose CHEESE, the first large-scale PCG dataset containing 24K portrait collections and 573K samples with high-quality modification text annotations, constructed through an Large Vison-Language Model-based pipeline with inversion-based verification. We further propose SCheese, a framework that combines text-guided generation with hierarchical identity and detail preservation. SCheese employs adaptive feature fusion mechanism to maintain identity consistency, and ConsistencyNet to inject fine-grained features for detail consistency. Comprehensive experiments validate the effectiveness of CHEESE in advancing PCG, with SCheese achieving state-of-the-art performance in handling complex edits with identity and fine-grained details consistency.
Acceleration methods for diffusion models (e.g., token merging or downsampling) typically optimize for synthesis quality under reduced compute, yet they often ignore the model's latent discriminative capacity. We revisit token compression with a joint objective and present **BiGain**, a training-free, plug-and-play framework that preserves generation quality while markedly improving classification in accelerated diffusion models. Our key insight is frequency separation: mapping feature-space signals into a frequency-aware representation disentangles fine detail from global semantics, enabling compression that respects both generative fidelity and discriminative utility. BiGain reflects this principle with two frequency-aware operators: (1) *Laplacian-gated token merging*, which encourages merges among spectrally smooth tokens while discouraging merges of high-contrast tokens, thereby retaining edges and textures; and (2) *Interpolate-Extrapolate KV Downsampling*, which downsamples keys/values via a controllable interextrapolation between nearest and average pooling while keeping queries intact, thereby conserving attention precision without retraining. Across DiT- and U-Net-based backbones and multiple datasets of ImageNet-1K, ImageNet-100, Oxford-IIIT Pets, and COCO-2017, our proposed operators consistently improve the speed-accuracy trade-off for diffusion-based classification, while maintaining, sometimes even enhancing generation quality under comparable acceleration. For instance, on ImageNet-1K, with 70% token merging ratio on Stable Diffusion 2.0, BiGain increases classification accuracy by 7.15% while also improving FID for generation by 0.34 (1.85%). Our comprehensive analyses indicate that balanced spectral retention, preserving high-frequency detail alongside low/mid-frequency semantic content is a reliable design rule for token compression in diffusion models. To our knowledge, BiGain is the first framework to jointly study and advance both generation and classification under accelerated diffusion, supporting lower-cost deployment of dual-purpose generative systems.
Learned world models excel at interpolative generalization but fail at extrapolative generalization to novel physical properties. This limitation arises because they learn statistical correlations rather than the environment's underlying generative rules, such as physical invariances and conservation laws. We argue that learning these invariances is key to robust extrapolation. To achieve this, we first introduce Symmetry Exploration, an unsupervised exploration strategy where an agent is intrinsically motivated by a Hamiltonian-based curiosity bonus to actively probe and challenge its understanding of conservation laws, thereby collecting physically informative data. Second, we design a Hamiltonian-based world model that learns from the collected data, using a novel self-supervised contrastive objective to identify the invariant physical state from raw, view-dependent pixel observations. Our framework, DreamSAC, trained on this actively curated data, significantly outperforms state-of-the-art baselines in 3D physics simulations on tasks requiring extrapolation.
BeautyGRPO: Aesthetic Alignment for Face Retouching via Dynamic Path Guidance and Fine-Grained Preference Modeling
PDF ↗Face retouching requires removing subtle imperfections while preserving unique facial identity features, in order to enhance overall aesthetic appeal. However, existing methods suffer from a fundamental trade-off. Supervised learning on labeled data is constrained to pixel-level label mimicry, failing to capture complex subjective human aesthetic preferences. Conversely, while online reinforcement learning (RL) excels at preference alignment, its stochastic exploration paradigm conflicts with the high-fidelity demands of face retouching and often introduces noticeable noise artifacts due to accumulated stochastic drift. To address these limitations, we propose BeautyGRPO, a reinforcement learning framework that aligns face retouching with human aesthetic preferences. We construct FRPref-10K, a fine-grained preference dataset covering five key retouching dimensions, and train a specialized reward model capable of evaluating subtle perceptual differences. To reconcile exploration and fidelity, we introduce Dynamic Path Guidance (DPG). DPG stabilizes the stochastic sampling trajectory by dynamically computing an anchor-based ODE path and replanning a guided trajectory at each sampling timestep, effectively correcting stochastic drift while maintaining controlled exploration. Extensive experiments show that BeautyGRPO outperforms both specialized face retouching methods and general image editing models, achieving superior texture quality, more accurate blemish removal, and overall results that better align with human aesthetic preferences. Our code will be made publicly available.
Model editing aims to update knowledge to add new concepts and change relevant information without retraining. Lifelong editing is a challenging task, prone to disrupting previously learned concepts, especially for Vision Language Models (VLMs), because sequential edits can lead to degraded reasoning and cross-modal misalignment. Existing VLM knowledge editing methods based on gated adapters, activation edits, and parameter merging techniques address catastrophic forgetting seen in full fine-tuning; however, they still operate in the shared representation space of the VLM, where concepts are entangled, so edits interfere with other non-relevant concepts. We hypothesize that this instability persists because current methods algorithmically control edits via optimization rather than structurally separating knowledge. We introduce Dynamic Subspace Concept Alignment (DSCA) which by design mitigates this limitation by decomposing the representation space into a set of orthogonal semantic subspaces and proposing edits only in those transformed spaces. These subspaces are obtained through incremental clustering and PCA on joint vision-language representations. This process structurally isolates concepts, enabling precise, non-interfering edits by turning isolation from a soft training objective into an architectural property. The surgical edits are guided by a multi-term loss function for maintaining task fidelity, edit locality, and cross-modal alignment. With the base model frozen, our method achieves 98% single-edit success, remains over 95% after 1,000 sequential edits, lowers hallucination by 3-5%, and achieves the best backward transfer (BWT) scores on continual instruction-tuning benchmarks. Extensive experiments demonstrate DSCA's state-of-the-art stability and knowledge retention capability in continual lifelong editing across various datasets and benchmarks.
Existing approaches to 3D semantic urban scene generation predominantly rely on voxel-based representations, which are bound by fixed resolution, challenging to edit, and memory-intensive in their dense form. In contrast, we advocate for a primitive-based paradigm where urban scenes are represented using compact, semantically meaningful 3D elements that are easy to manipulate and compose. To this end, we introduce PrITTI, a latent diffusion model that leverages vectorized object primitives and rasterized ground surfaces for generating diverse, controllable, and editable 3D semantic urban scenes. This hybrid representation yields a structured latent space that facilitates object- and ground-level manipulation. Experiments on KITTI-360 show that primitive-based representations unlock the full capabilities of diffusion transformers, achieving state-of-the-art 3D scene generation quality with lower memory requirements, faster inference, and greater editability than voxel-based methods. Beyond generation, PrITTI supports a range of downstream applications, including scene editing, inpainting, outpainting, and photo-realistic street-view synthesis.
Moving instance segmentation (MIS) attracts increasing attention due to its broad applications in traffic surveillance, autonomous driving, and animal tracking. Event cameras record asynchronous brightness changes, providing high temporal resolution and dynamic range, which makes them highly sensitive to motion information. By fusing event and image features, motion cues from events can complement spatial details from images, enhancing the performance of MIS. However, current multimodal MIS methods still struggle to segment small moving instances, as event cameras often yield sparse features under limited resolution. Moreover, event features entangle appearance attributes with motion cues, which further restricts effective cross-modal fusion. To address these challenges, we first propose a dual-disentangling feature extraction framework that separates and extracts appearance and motion information within both image and event modalities, thereby improving feature density. Subsequently, a multi-granularity cross-modal alignment is introduced to align distributionally and semantically consistent features across modalities, enabling more effective fusion with rich spatial and temporal details. The experiment results demonstrate that our method achieves state-of-the-art performance in multimodal MIS, especially for small instances under challenging conditions such as fast motion and low-light settings.
Despite recent progress, reinforcement learning (RL)-based fine-tuning of diffusion models often struggles with generalization, composability, and robustness against reward hacking. Recent studies have explored prompt refinement as a modular alternative, but most adopt a feed-forward approach that applies a single refined prompt throughout the entire sampling trajectory, thereby failing to fully leverage the sequential nature of reinforcement learning. To address this, we introduce PromptLoop, a plug-and-play RL framework that incorporates latent feedback into step-wise prompt refinement. Rather than modifying diffusion model weights, a multimodal large language model (MLLM) is trained with RL to iteratively update prompts based on intermediate latent states of diffusion models. This design achieves a structural analogy to the Diffusion RL approach, while retaining the flexibility and generality of prompt-based alignment. Extensive experiments across diverse reward functions and diffusion backbones demonstrate that PromptLoop (i) achieves effective reward optimization, (ii) generalizes seamlessly to unseen models, (iii) composes orthogonally with existing alignment methods, and (iv) mitigates over-optimization and reward hacking while introducing only a practically negligible inference overhead.