Kehan Li

Jun Cen, Siteng Huang, Yuqian Yuan et al. · pku

We introduce RynnVLA-002, a unified Vision-Language-Action (VLA) and world model. The world model leverages action and visual inputs to predict future image states, learning the underlying physics of the environment to refine action generation. Conversely, the VLA model produces subsequent actions from image observations, enhancing visual understanding and supporting the world model's image generation. The unified framework of RynnVLA-002 enables joint learning of environmental dynamics and action planning. Our experiments show that RynnVLA-002 surpasses individual VLA and world models, demonstrating their mutual enhancement. We evaluate RynnVLA-002 in both simulation and real-world robot tasks. RynnVLA-002 achieves 97.4% success rate on the LIBERO simulation benchmark without pretraining, while in real-world LeRobot experiments, its integrated world model boosts the overall success rate by 50%.

Peng Jin, Hao Li, Zesen Cheng et al. · pku

Existing text-video retrieval solutions are, in essence, discriminant models focused on maximizing the conditional likelihood, i.e., p(candidates|query). While straightforward, this de facto paradigm overlooks the underlying data distribution p(query), which makes it challenging to identify out-of-distribution data. To address this limitation, we creatively tackle this task from a generative viewpoint and model the correlation between the text and the video as their joint probability p(candidates,query). This is accomplished through a diffusion-based text-video retrieval framework (DiffusionRet), which models the retrieval task as a process of gradually generating joint distribution from noise. During training, DiffusionRet is optimized from both the generation and discrimination perspectives, with the generator being optimized by generation loss and the feature extractor trained with contrastive loss. In this way, DiffusionRet cleverly leverages the strengths of both generative and discriminative methods. Extensive experiments on five commonly used text-video retrieval benchmarks, including MSRVTT, LSMDC, MSVD, ActivityNet Captions, and DiDeMo, with superior performances, justify the efficacy of our method. More encouragingly, without any modification, DiffusionRet even performs well in out-domain retrieval settings. We believe this work brings fundamental insights into the related fields. Code is available at https://github.com/jpthu17/DiffusionRet.

Xiaoyu Lin, Nicholas Tagliapietra, Kehan Li et al.

Root-Cause Analysis (RCA) seeks to identify the variables responsible for abnormal system behavior in complex domains such as manufacturing, cloud computing, and healthcare. Existing approaches face a critical bottleneck: graph-based causal methods can identify intervention targets but typically require a known or accurately estimated causal graph, while graph-free statistical methods either localize marginal anomalies rather than structural causes, or rely on restrictive assumptions about graph structure or functional form. We propose StableRCA, a local mechanism-level RCA framework that avoids global graph discovery by estimating local Markov boundaries and detecting conditional distribution shifts within them. Leveraging the Independent Causal Mechanism principle, we show that intervention targets can be identified with probability converging exponentially in sample size under faithful Markov boundary recovery and non-degenerate mechanism shifts. Experiments on synthetic benchmarks and five real-world datasets demonstrate that StableRCA is robust to graph misspecification, effective under multiple intervention targets, scalable to large systems, and reliable across diverse application domains. Code is available at: https://anonymous.4open.science/r/StableRCA-E362

Kehan Li, Runyi Yu, Zhennan Wang et al. · pku

While the Vision Transformer (VT) architecture is becoming trendy in computer vision, pure VT models perform poorly on tiny datasets. To address this issue, this paper proposes the locality guidance for improving the performance of VTs on tiny datasets. We first analyze that the local information, which is of great importance for understanding images, is hard to be learned with limited data due to the high flexibility and intrinsic globality of the self-attention mechanism in VTs. To facilitate local information, we realize the locality guidance for VTs by imitating the features of an already trained convolutional neural network (CNN), inspired by the built-in local-to-global hierarchy of CNN. Under our dual-task learning paradigm, the locality guidance provided by a lightweight CNN trained on low-resolution images is adequate to accelerate the convergence and improve the performance of VTs to a large extent. Therefore, our locality guidance approach is very simple and efficient, and can serve as a basic performance enhancement method for VTs on tiny datasets. Extensive experiments demonstrate that our method can significantly improve VTs when training from scratch on tiny datasets and is compatible with different kinds of VTs and datasets. For example, our proposed method can boost the performance of various VTs on tiny datasets (e.g., 13.07% for DeiT, 8.98% for T2T and 7.85% for PVT), and enhance even stronger baseline PVTv2 by 1.86% to 79.30%, showing the potential of VTs on tiny datasets. The code is available at https://github.com/lkhl/tiny-transformers.

Runyi Yu, Zhennan Wang, Yinhuai Wang et al. · pku

The Position Embedding (PE) is critical for Vision Transformers (VTs) due to the permutation-invariance of self-attention operation. By analyzing the input and output of each encoder layer in VTs using reparameterization and visualization, we find that the default PE joining method (simply adding the PE and patch embedding together) operates the same affine transformation to token embedding and PE, which limits the expressiveness of PE and hence constrains the performance of VTs. To overcome this limitation, we propose a simple, effective, and robust method. Specifically, we provide two independent layer normalizations for token embeddings and PE for each layer, and add them together as the input of each layer's Muti-Head Self-Attention module. Since the method allows the model to adaptively adjust the information of PE for different layers, we name it as Layer-adaptive Position Embedding, abbreviated as LaPE. Extensive experiments demonstrate that LaPE can improve various VTs with different types of PE and make VTs robust to PE types. For example, LaPE improves 0.94% accuracy for ViT-Lite on Cifar10, 0.98% for CCT on Cifar100, and 1.72% for DeiT on ImageNet-1K, which is remarkable considering the negligible extra parameters, memory and computational cost brought by LaPE. The code is publicly available at https://github.com/Ingrid725/LaPE.

Zesen Cheng, Peng Jin, Hao Li et al. · pku

The top-down and bottom-up methods are two mainstreams of referring segmentation, while both methods have their own intrinsic weaknesses. Top-down methods are chiefly disturbed by Polar Negative (PN) errors owing to the lack of fine-grained cross-modal alignment. Bottom-up methods are mainly perturbed by Inferior Positive (IP) errors due to the lack of prior object information. Nevertheless, we discover that two types of methods are highly complementary for restraining respective weaknesses but the direct average combination leads to harmful interference. In this context, we build Win-win Cooperation (WiCo) to exploit complementary nature of two types of methods on both interaction and integration aspects for achieving a win-win improvement. For the interaction aspect, Complementary Feature Interaction (CFI) provides fine-grained information to top-down branch and introduces prior object information to bottom-up branch for complementary feature enhancement. For the integration aspect, Gaussian Scoring Integration (GSI) models the gaussian performance distributions of two branches and weightedly integrates results by sampling confident scores from the distributions. With our WiCo, several prominent top-down and bottom-up combinations achieve remarkable improvements on three common datasets with reasonable extra costs, which justifies effectiveness and generality of our method.

Chen Shen, Jun Zhang, Xinggong Liang et al.

Forensic pathology is critical in analyzing death manner and time from the microscopic aspect to assist in the establishment of reliable factual bases for criminal investigation. In practice, even the manual differentiation between different postmortem organ tissues is challenging and relies on expertise, considering that changes like putrefaction and autolysis could significantly change typical histopathological appearance. Developing AI-based computational pathology techniques to assist forensic pathologists is practically meaningful, which requires reliable discriminative representation learning to capture tissues' fine-grained postmortem patterns. To this end, we propose a framework called FPath, in which a dedicated self-supervised contrastive learning strategy and a context-aware multiple-instance learning (MIL) block are designed to learn discriminative representations from postmortem histopathological images acquired at varying magnification scales. Our self-supervised learning step leverages multiple complementary contrastive losses and regularization terms to train a double-tier backbone for fine-grained and informative patch/instance embedding. Thereafter, the context-aware MIL adaptively distills from the local instances a holistic bag/image-level representation for the recognition task. On a large-scale database of $19,607$ experimental rat postmortem images and $3,378$ real-world human decedent images, our FPath led to state-of-the-art accuracy and promising cross-domain generalization in recognizing seven different postmortem tissues. The source code will be released on \href{https://github.com/ladderlab-xjtu/forensic_pathology}{https://github.com/ladderlab-xjtu/forensic\_pathology}.

Zesen Cheng, Pengchong Qiao, Kehan Li et al. · pku

Weakly supervised semantic segmentation is typically inspired by class activation maps, which serve as pseudo masks with class-discriminative regions highlighted. Although tremendous efforts have been made to recall precise and complete locations for each class, existing methods still commonly suffer from the unsolicited Out-of-Candidate (OC) error predictions that not belongs to the label candidates, which could be avoidable since the contradiction with image-level class tags is easy to be detected. In this paper, we develop a group ranking-based Out-of-Candidate Rectification (OCR) mechanism in a plug-and-play fashion. Firstly, we adaptively split the semantic categories into In-Candidate (IC) and OC groups for each OC pixel according to their prior annotation correlation and posterior prediction correlation. Then, we derive a differentiable rectification loss to force OC pixels to shift to the IC group. Incorporating our OCR with seminal baselines (e.g., AffinityNet, SEAM, MCTformer), we can achieve remarkable performance gains on both Pascal VOC (+3.2%, +3.3%, +0.8% mIoU) and MS COCO (+1.0%, +1.3%, +0.5% mIoU) datasets with negligible extra training overhead, which justifies the effectiveness and generality of our OCR.

Kehan Li, Zhennan Wang, Zesen Cheng et al. · pku

Recently, self-supervised large-scale visual pre-training models have shown great promise in representing pixel-level semantic relationships, significantly promoting the development of unsupervised dense prediction tasks, e.g., unsupervised semantic segmentation (USS). The extracted relationship among pixel-level representations typically contains rich class-aware information that semantically identical pixel embeddings in the representation space gather together to form sophisticated concepts. However, leveraging the learned models to ascertain semantically consistent pixel groups or regions in the image is non-trivial since over/ under-clustering overwhelms the conceptualization procedure under various semantic distributions of different images. In this work, we investigate the pixel-level semantic aggregation in self-supervised ViT pre-trained models as image Segmentation and propose the Adaptive Conceptualization approach for USS, termed ACSeg. Concretely, we explicitly encode concepts into learnable prototypes and design the Adaptive Concept Generator (ACG), which adaptively maps these prototypes to informative concepts for each image. Meanwhile, considering the scene complexity of different images, we propose the modularity loss to optimize ACG independent of the concept number based on estimating the intensity of pixel pairs belonging to the same concept. Finally, we turn the USS task into classifying the discovered concepts in an unsupervised manner. Extensive experiments with state-of-the-art results demonstrate the effectiveness of the proposed ACSeg.

Zesen Cheng, Kehan Li, Peng Jin et al. · pku

Unified visual grounding pursues a simple and generic technical route to leverage multi-task data with less task-specific design. The most advanced methods typically present boxes and masks as vertex sequences to model referring detection and segmentation as an autoregressive sequential vertex generation paradigm. However, generating high-dimensional vertex sequences sequentially is error-prone because the upstream of the sequence remains static and cannot be refined based on downstream vertex information, even if there is a significant location gap. Besides, with limited vertexes, the inferior fitting of objects with complex contours restricts the performance upper bound. To deal with this dilemma, we propose a parallel vertex generation paradigm for superior high-dimension scalability with a diffusion model by simply modifying the noise dimension. An intuitive materialization of our paradigm is Parallel Vertex Diffusion (PVD) to directly set vertex coordinates as the generation target and use a diffusion model to train and infer. We claim that it has two flaws: (1) unnormalized coordinate caused a high variance of loss value; (2) the original training objective of PVD only considers point consistency but ignores geometry consistency. To solve the first flaw, Center Anchor Mechanism (CAM) is designed to convert coordinates as normalized offset values to stabilize the training loss value. For the second flaw, Angle summation loss (ASL) is designed to constrain the geometry difference of prediction and ground truth vertexes for geometry-level consistency. Empirical results show that our PVD achieves state-of-the-art in both referring detection and segmentation, and our paradigm is more scalable and efficient than sequential vertex generation with high-dimension data.

Peng Jin, Hao Li, Zesen Cheng et al. · pku

Text-to-motion generation requires not only grounding local actions in language but also seamlessly blending these individual actions to synthesize diverse and realistic global motions. However, existing motion generation methods primarily focus on the direct synthesis of global motions while neglecting the importance of generating and controlling local actions. In this paper, we propose the local action-guided motion diffusion model, which facilitates global motion generation by utilizing local actions as fine-grained control signals. Specifically, we provide an automated method for reference local action sampling and leverage graph attention networks to assess the guiding weight of each local action in the overall motion synthesis. During the diffusion process for synthesizing global motion, we calculate the local-action gradient to provide conditional guidance. This local-to-global paradigm reduces the complexity associated with direct global motion generation and promotes motion diversity via sampling diverse actions as conditions. Extensive experiments on two human motion datasets, i.e., HumanML3D and KIT, demonstrate the effectiveness of our method. Furthermore, our method provides flexibility in seamlessly combining various local actions and continuous guiding weight adjustment, accommodating diverse user preferences, which may hold potential significance for the community. The project page is available at https://jpthu17.github.io/GuidedMotion-project/.

Kehan Li, Yian Zhao, Zhennan Wang et al. · pku

Interactive segmentation enables users to segment as needed by providing cues of objects, which introduces human-computer interaction for many fields, such as image editing and medical image analysis. Typically, massive and expansive pixel-level annotations are spent to train deep models by object-oriented interactions with manually labeled object masks. In this work, we reveal that informative interactions can be made by simulation with semantic-consistent yet diverse region exploration in an unsupervised paradigm. Concretely, we introduce a Multi-granularity Interaction Simulation (MIS) approach to open up a promising direction for unsupervised interactive segmentation. Drawing on the high-quality dense features produced by recent self-supervised models, we propose to gradually merge patches or regions with similar features to form more extensive regions and thus, every merged region serves as a semantic-meaningful multi-granularity proposal. By randomly sampling these proposals and simulating possible interactions based on them, we provide meaningful interaction at multiple granularities to teach the model to understand interactions. Our MIS significantly outperforms non-deep learning unsupervised methods and is even comparable with some previous deep-supervised methods without any annotation.

Zhennan Wang, Kehan Li, Runyi Yu et al. · pku

In this paper, we analyze batch normalization from the perspective of discriminability and find the disadvantages ignored by previous studies: the difference in $l_2$ norms of sample features can hinder batch normalization from obtaining more distinguished inter-class features and more compact intra-class features. To address this issue, we propose a simple yet effective method to equalize the $l_2$ norms of sample features. Concretely, we $l_2$-normalize each sample feature before feeding them into batch normalization, and therefore the features are of the same magnitude. Since the proposed method combines the $l_2$ normalization and batch normalization, we name our method $L_2$BN. The $L_2$BN can strengthen the compactness of intra-class features and enlarge the discrepancy of inter-class features. The $L_2$BN is easy to implement and can exert its effect without any additional parameters or hyper-parameters. We evaluate the effectiveness of $L_2$BN through extensive experiments with various models on image classification and acoustic scene classification tasks. The results demonstrate that the $L_2$BN can boost the generalization ability of various neural network models and achieve considerable performance improvements.

Zeyu Jin, Jia Jia, Qixin Wang et al.

Speech-language multi-modal learning presents a significant challenge due to the fine nuanced information inherent in speech styles. Therefore, a large-scale dataset providing elaborate comprehension of speech style is urgently needed to facilitate insightful interplay between speech audio and natural language. However, constructing such datasets presents a major trade-off between large-scale data collection and high-quality annotation. To tackle this challenge, we propose an automatic speech annotation system for expressiveness interpretation that annotates in-the-wild speech clips with expressive and vivid human language descriptions. Initially, speech audios are processed by a series of expert classifiers and captioning models to capture diverse speech characteristics, followed by a fine-tuned LLaMA for customized annotation generation. Unlike previous tag/templet-based annotation frameworks with limited information and diversity, our system provides in-depth understandings of speech style through tailored natural language descriptions, thereby enabling accurate and voluminous data generation for large model training. With this system, we create SpeechCraft, a fine-grained bilingual expressive speech dataset. It is distinguished by highly descriptive natural language style prompts, containing approximately 2,000 hours of audio data and encompassing over two million speech clips. Extensive experiments demonstrate that the proposed dataset significantly boosts speech-language task performance in stylist speech synthesis and speech style understanding.

Chen Shen, Chunfeng Lian, Wanqing Zhang et al.

Forensic pathology is critical in determining the cause and manner of death through post-mortem examinations, both macroscopic and microscopic. The field, however, grapples with issues such as outcome variability, laborious processes, and a scarcity of trained professionals. This paper presents SongCi, an innovative visual-language model (VLM) designed specifically for forensic pathology. SongCi utilizes advanced prototypical cross-modal self-supervised contrastive learning to enhance the accuracy, efficiency, and generalizability of forensic analyses. It was pre-trained and evaluated on a comprehensive multi-center dataset, which includes over 16 million high-resolution image patches, 2,228 vision-language pairs of post-mortem whole slide images (WSIs), and corresponding gross key findings, along with 471 distinct diagnostic outcomes. Our findings indicate that SongCi surpasses existing multi-modal AI models in many forensic pathology tasks, performs comparably to experienced forensic pathologists and significantly better than less experienced ones, and provides detailed multi-modal explainability, offering critical assistance in forensic investigations. To the best of our knowledge, SongCi is the first VLM specifically developed for forensic pathological analysis and the first large-vocabulary computational pathology (CPath) model that directly processes gigapixel WSIs in forensic science.

Han Yu, Hao Zou, Xingxuan Zhang et al.

Although neural networks achieve promising performance in many tasks, they may still fail when encountering some examples and bring about risks to applications. To discover risky samples, previous literature attempts to search for patterns of risky samples within existing datasets or inject perturbation into them. Yet in this way the diversity of risky samples is limited by the coverage of existing datasets. To overcome this limitation, recent works adopt diffusion models to produce new risky samples beyond the coverage of existing datasets. However, these methods struggle in the conformity between generated samples and expected categories, which could introduce label noise and severely limit their effectiveness in applications. To address this issue, we propose RiskyDiff that incorporates the embeddings of both texts and images as implicit constraints of category conformity. We also design a conformity score to further explicitly strengthen the category conformity, as well as introduce the mechanisms of embedding screening and risky gradient guidance to boost the risk of generated samples. Extensive experiments reveal that RiskyDiff greatly outperforms existing methods in terms of the degree of risk, generation quality, and conformity with conditioned categories. We also empirically show the generalization ability of the models can be enhanced by augmenting training data with generated samples of high conformity.

Han Yu, Kehan Li, Dongbai Li et al.

Recently, there has been gradually more attention paid to Out-of-Distribution (OOD) performance prediction, whose goal is to predict the performance of trained models on unlabeled OOD test datasets, so that we could better leverage and deploy off-the-shelf trained models in risk-sensitive scenarios. Although progress has been made in this area, evaluation protocols in previous literature are inconsistent, and most works cover only a limited number of real-world OOD datasets and types of distribution shifts. To provide convenient and fair comparisons for various algorithms, we propose Out-of-Distribution Performance Prediction Benchmark (ODP-Bench), a comprehensive benchmark that includes most commonly used OOD datasets and existing practical performance prediction algorithms. We provide our trained models as a testbench for future researchers, thus guaranteeing the consistency of comparison and avoiding the burden of repeating the model training process. Furthermore, we also conduct in-depth experimental analyses to better understand their capability boundary.

Ronghao Dang, Yuqian Yuan, Yunxuan Mao et al.

We introduce RynnEC, a video multimodal large language model designed for embodied cognition. Built upon a general-purpose vision-language foundation model, RynnEC incorporates a region encoder and a mask decoder, enabling flexible region-level video interaction. Despite its compact architecture, RynnEC achieves state-of-the-art performance in object property understanding, object segmentation, and spatial reasoning. Conceptually, it offers a region-centric video paradigm for the brain of embodied agents, providing fine-grained perception of the physical world and enabling more precise interactions. To mitigate the scarcity of annotated 3D datasets, we propose an egocentric video based pipeline for generating embodied cognition data. Furthermore, we introduce RynnEC-Bench, a region-centered benchmark for evaluating embodied cognitive capabilities. We anticipate that RynnEC will advance the development of general-purpose cognitive cores for embodied agents and facilitate generalization across diverse embodied tasks. The code, model checkpoints, and benchmark are available at: https://github.com/alibaba-damo-academy/RynnEC

Wenwen Tong, Hewei Guo, Dongchuan Ran et al.

We introduce InteractiveOmni, a unified and open-source omni-modal large language model for audio-visual multi-turn interaction, ranging from 4B to 8B parameters, designed to lead the field of lightweight models by offering comprehensive omni-modal understanding and speech generation capabilities. To achieve this, we integrate the vision encoder, audio encoder, large language model, and speech decoder into a unified model for understanding and generation tasks. We design a multi-stage training strategy to ensure robust cross-modal capabilities, including pre-training for omni-modal understanding, followed by post-training with speech conversation and audio-visual interaction. To enable human-like long-term conversational ability, we meticulously curate a multi-turn training dataset that enhances the model's ability to handle complex and multi-turn interactions. To effectively evaluate the multi-turn memory and speech interaction capabilities, we construct the multi-modal multi-turn memory benchmark and the multi-turn speech interaction benchmark. Experiments demonstrate that InteractiveOmni significantly outperforms leading open-source models and provides a more intelligent multi-turn audio-visual experience, particularly in its long-term memory capabilities. Notably, InteractiveOmni-4B is comparable to the much larger model like Qwen2.5-Omni-7B on general benchmarks, and it can retain 97% of the performance of the InteractiveOmni-8B while utilizing only 50% of the model size. Achieving state-of-the-art results against similarly sized models across image, audio, video understanding, and speech generation tasks, InteractiveOmni is an accessible, open-source foundation for next-generation intelligent interactive systems.

Boqiang Zhang, Kehan Li, Zesen Cheng et al. · pku

In this paper, we propose VideoLLaMA3, a more advanced multimodal foundation model for image and video understanding. The core design philosophy of VideoLLaMA3 is vision-centric. The meaning of "vision-centric" is two-fold: the vision-centric training paradigm and vision-centric framework design. The key insight of our vision-centric training paradigm is that high-quality image-text data is crucial for both image and video understanding. Instead of preparing massive video-text datasets, we focus on constructing large-scale and high-quality image-text datasets. VideoLLaMA3 has four training stages: 1) Vision Encoder Adaptation, which enables vision encoder to accept images of variable resolutions as input; 2) Vision-Language Alignment, which jointly tunes the vision encoder, projector, and LLM with large-scale image-text data covering multiple types (including scene images, documents, charts) as well as text-only data. 3) Multi-task Fine-tuning, which incorporates image-text SFT data for downstream tasks and video-text data to establish a foundation for video understanding. 4) Video-centric Fine-tuning, which further improves the model's capability in video understanding. As for the framework design, to better capture fine-grained details in images, the pretrained vision encoder is adapted to encode images of varying sizes into vision tokens with corresponding numbers, rather than a fixed number of tokens. For video inputs, we reduce the number of vision tokens according to their similarity so that the representation of videos will be more precise and compact. Benefit from vision-centric designs, VideoLLaMA3 achieves compelling performances in both image and video understanding benchmarks.

Yanyun Pu, Kehan Li, Zeyi Huang et al.

With the rapid advancement of video generation models such as Sora, video quality assessment (VQA) is becoming increasingly crucial for selecting high-quality videos from large-scale datasets used in pre-training. Traditional VQA methods, typically producing single numerical scores, often lack comprehensiveness and interpretability. To address these challenges, we introduce MVQA-68K, a novel multi-dimensional VQA dataset comprising over 68,000 carefully annotated videos, covering seven essential quality dimensions: overall aesthetics, camera movement, dynamic degree, texture detail, composition, visual quality, and factual consistency. Each annotation includes detailed chain-of-thought reasoning to facilitate interpretability and comprehensive understanding. Extensive experiments demonstrate that MVQA-68K significantly enhances the performance of various multimodal large language models (MLLMs) on the VQA task, achieving state-of-the-art results not only on our internal test set (Fig.1) but also on public benchmarks including LSVQ-test, LSVQ-1080p, and LIVE-VQC. Meantime, incorporating explicit reasoning process during VQA training substantially boosts the zero-shot generalization. Code and dataset will be available at github: https://github.com/Controller01-ai/MVQA-68K

Zhizhong Wang, Tianyi Chu, Zeyi Huang et al.

Personalized Text-to-Image (PT2I) generation aims to produce customized images based on reference images. A prominent interest pertains to the integration of an image prompt adapter to facilitate zero-shot PT2I without test-time fine-tuning. However, current methods grapple with three fundamental challenges: 1. the elusive equilibrium between Concept Preservation (CP) and Prompt Following (PF), 2. the difficulty in retaining fine-grained concept details in reference images, and 3. the restricted scalability to extend to multi-subject personalization. To tackle these challenges, we present Dynamic Image Prompt Adapter (DynaIP), a cutting-edge plugin to enhance the fine-grained concept fidelity, CP-PF balance, and subject scalability of SOTA T2I multimodal diffusion transformers (MM-DiT) for PT2I generation. Our key finding is that MM-DiT inherently exhibit decoupling learning behavior when injecting reference image features into its dual branches via cross attentions. Based on this, we design an innovative Dynamic Decoupling Strategy that removes the interference of concept-agnostic information during inference, significantly enhancing the CP-PF balance and further bolstering the scalability of multi-subject compositions. Moreover, we identify the visual encoder as a key factor affecting fine-grained CP and reveal that the hierarchical features of commonly used CLIP can capture visual information at diverse granularity levels. Therefore, we introduce a novel Hierarchical Mixture-of-Experts Feature Fusion Module to fully leverage the hierarchical features of CLIP, remarkably elevating the fine-grained concept fidelity while also providing flexible control of visual granularity. Extensive experiments across single- and multi-subject PT2I tasks verify that our DynaIP outperforms existing approaches, marking a notable advancement in the field of PT2l generation.

Hao Li, Curise Jia, Peng Jin et al. · pku

Image Retrieval aims to retrieve corresponding images based on a given query. In application scenarios, users intend to express their retrieval intent through various query styles. However, current retrieval tasks predominantly focus on text-query retrieval exploration, leading to limited retrieval query options and potential ambiguity or bias in user intention. In this paper, we propose the Style-Diversified Query-Based Image Retrieval task, which enables retrieval based on various query styles. To facilitate the novel setting, we propose the first Diverse-Style Retrieval dataset, encompassing diverse query styles including text, sketch, low-resolution, and art. We also propose a light-weighted style-diversified retrieval framework. For various query style inputs, we apply the Gram Matrix to extract the query's textural features and cluster them into a style space with style-specific bases. Then we employ the style-init prompt tuning module to enable the visual encoder to comprehend the texture and style information of the query. Experiments demonstrate that our model, employing the style-init prompt tuning strategy, outperforms existing retrieval models on the style-diversified retrieval task. Moreover, style-diversified queries~(sketch+text, art+text, etc) can be simultaneously retrieved in our model. The auxiliary information from other queries enhances the retrieval performance within the respective query.

Yian Zhao, Kehan Li, Zesen Cheng et al. · pku

Interactive Segmentation (IS) segments specific objects or parts in the image according to user input. Current IS pipelines fall into two categories: single-granularity output and multi-granularity output. The latter aims to alleviate the spatial ambiguity present in the former. However, the multi-granularity output pipeline suffers from limited interaction flexibility and produces redundant results. In this work, we introduce Granularity-Controllable Interactive Segmentation (GraCo), a novel approach that allows precise control of prediction granularity by introducing additional parameters to input. This enhances the customization of the interactive system and eliminates redundancy while resolving ambiguity. Nevertheless, the exorbitant cost of annotating multi-granularity masks and the lack of available datasets with granularity annotations make it difficult for models to acquire the necessary guidance to control output granularity. To address this problem, we design an any-granularity mask generator that exploits the semantic property of the pre-trained IS model to automatically generate abundant mask-granularity pairs without requiring additional manual annotation. Based on these pairs, we propose a granularity-controllable learning strategy that efficiently imparts the granularity controllability to the IS model. Extensive experiments on intricate scenarios at object and part levels demonstrate that our GraCo has significant advantages over previous methods. This highlights the potential of GraCo to be a flexible annotation tool, capable of adapting to diverse segmentation scenarios. The project page: https://zhao-yian.github.io/GraCo.

XuDong Wang, Shaolun Zhang, Shufan Li et al.

We present SegLLM, a novel multi-round interactive reasoning segmentation model that enhances LLM-based segmentation by exploiting conversational memory of both visual and textual outputs. By leveraging a mask-aware multimodal LLM, SegLLM re-integrates previous segmentation results into its input stream, enabling it to reason about complex user intentions and segment objects in relation to previously identified entities, including positional, interactional, and hierarchical relationships, across multiple interactions. This capability allows SegLLM to respond to visual and text queries in a chat-like manner. Evaluated on the newly curated MRSeg benchmark, SegLLM outperforms existing methods in multi-round interactive reasoning segmentation by over 20%. Additionally, we observed that training on multi-round reasoning segmentation data enhances performance on standard single-round referring segmentation and localization tasks, resulting in a 5.5% increase in cIoU for referring expression segmentation and a 4.5% improvement in Acc@0.5 for referring expression localization.

Liyu Zhang, Kehan Li, Tingrui Han et al.

Post training via GRPO has demonstrated remarkable effectiveness in improving the generation quality of flow-matching models. However, GRPO suffers from inherently low sample efficiency due to its on-policy training paradigm. To address this limitation, we present OP-GRPO, the first Off-Policy GRPO framework tailored for flow-matching models. First, we actively select high-quality trajectories and adaptively incorporate them into a replay buffer for reuse in subsequent training iterations. Second, to mitigate the distribution shift introduced by off-policy samples, we propose a sequence-level importance sampling correction that preserves the integrity of GRPO's clipping mechanism while ensuring stable policy updates. Third, we theoretically and empirically show that late denoising steps yield ill-conditioned off-policy ratios, and mitigate this by truncating trajectories at late steps. Across image and video generation benchmarks, OP-GRPO achieves comparable or superior performance to Flow-GRPO with only 34.2% of the training steps on average, yielding substantial gains in training efficiency while maintaining generation quality.

Xingxuan Zhang, Gang Ren, Han Yu et al.

We argue that progress toward general intelligence requires complementary foundation models grounded in language, the physical world, and structured data. This report presents LimiX-16M and LimiX-2M, two instantiations of our large structured-data models (LDMs). Both models treat structured data as a joint distribution over variables and missingness, thus capable of addressing a wide range of tabular tasks through query-based conditional prediction via a single model. They are pretrained using masked joint-distribution modeling with an episodic, context-conditional objective, supporting rapid, training-free adaptation at inference. We evaluate LimiX models across 11 large structured-data benchmarks with broad regimes of sample size, feature dimensionality, class number, categorical-to-numerical feature ratio, missingness, and sample-to-feature ratios. LimiX-16M consistently surpasses strong baselines, as shown in Figure 1 and Figure 2. The superiority holds across a wide range of tasks, such as classification, regression, missing value imputation, and data generation, often by substantial margins, while avoiding task-specific architectures or bespoke training per task. Notably, LimiX-2M delivers strong results under tight compute and memory budgets. We also present the first scaling law study for LDMs, revealing how data and model scaling jointly influence downstream performance and offering quantitative guidance for tabular foundation modeling. All LimiX models are publicly accessible under Apache 2.0.

Yuming Jiang, Siteng Huang, Shengke Xue et al. · pku

This paper presents RynnVLA-001, a vision-language-action(VLA) model built upon large-scale video generative pretraining from human demonstrations. We propose a novel two-stage pretraining methodology. The first stage, Ego-Centric Video Generative Pretraining, trains an Image-to-Video model on 12M ego-centric manipulation videos to predict future frames conditioned on an initial frame and a language instruction. The second stage, Human-Centric Trajectory-Aware Modeling, extends this by jointly predicting future keypoint trajectories, thereby effectively bridging visual frame prediction with action prediction. Furthermore, to enhance action representation, we propose ActionVAE, a variational autoencoder that compresses sequences of actions into compact latent embeddings, reducing the complexity of the VLA output space. When finetuned on the same downstream robotics datasets, RynnVLA-001 achieves superior performance over state-of-the-art baselines, demonstrating that the proposed pretraining strategy provides a more effective initialization for VLA models.

Zesen Cheng, Hang Zhang, Kehan Li et al. · pku

Contrastive loss is a powerful approach for representation learning, where larger batch sizes enhance performance by providing more negative samples to better distinguish between similar and dissimilar data. However, scaling batch sizes is constrained by the quadratic growth in GPU memory consumption, primarily due to the full instantiation of the similarity matrix. To address this, we propose a tile-based computation strategy that partitions the contrastive loss calculation into arbitrary small blocks, avoiding full materialization of the similarity matrix. Furthermore, we introduce a multi-level tiling strategy to leverage the hierarchical structure of distributed systems, employing ring-based communication at the GPU level to optimize synchronization and fused kernels at the CUDA core level to reduce I/O overhead. Experimental results show that the proposed method scales batch sizes to unprecedented levels. For instance, it enables contrastive training of a CLIP-ViT-L/14 model with a batch size of 4M or 12M using 8 or 32 A800 80GB without sacrificing any accuracy. Compared to SOTA memory-efficient solutions, it achieves a two-order-of-magnitude reduction in memory while maintaining comparable speed. The code will be made publicly available.

Yuqian Yuan, Wenqiao Zhang, Xin Li et al.

Multimodal large language models (MLLMs) have demonstrated strong general-purpose capabilities in open-world visual comprehension. However, most existing MLLMs primarily focus on holistic, scene-level understanding, often overlooking the need for fine-grained, object-centric reasoning. In this paper, we present PixelRefer, a unified region-level MLLM framework that enables advanced fine-grained understanding over user-specified regions across both images and videos. Motivated by the observation that LLM attention predominantly focuses on object-level tokens, we propose a Scale-Adaptive Object Tokenizer (SAOT) to generate compact and semantically rich object representations from free-form regions. Our analysis reveals that global visual tokens contribute mainly in early LLM layers, inspiring the design of PixelRefer-Lite, an efficient variant that employs an Object-Centric Infusion module to pre-fuse global context into object tokens. This yields a lightweight Object-Only Framework that substantially reduces computational cost while maintaining high semantic fidelity. To facilitate fine-grained instruction tuning, we curate PixelRefer-2.2M, a high-quality object-centric instruction dataset. Extensive experiments across a range of benchmarks validate that PixelRefer achieves leading performance with fewer training samples, while PixelRefer-Lite offers competitive accuracy with notable gains in efficiency.

Chen Shen, Wanqing Zhang, Kehan Li et al.

Forensic cause-of-death determination faces systemic challenges, including workforce shortages and diagnostic variability, particularly in high-volume systems like China's medicolegal infrastructure. We introduce FEAT (ForEnsic AgenT), a multi-agent AI framework that automates and standardizes death investigations through a domain-adapted large language model. FEAT's application-oriented architecture integrates: (i) a central Planner for task decomposition, (ii) specialized Local Solvers for evidence analysis, (iii) a Memory & Reflection module for iterative refinement, and (iv) a Global Solver for conclusion synthesis. The system employs tool-augmented reasoning, hierarchical retrieval-augmented generation, forensic-tuned LLMs, and human-in-the-loop feedback to ensure legal and medical validity. In evaluations across diverse Chinese case cohorts, FEAT outperformed state-of-the-art AI systems in both long-form autopsy analyses and concise cause-of-death conclusions. It demonstrated robust generalization across six geographic regions and achieved high expert concordance in blinded validations. Senior pathologists validated FEAT's outputs as comparable to those of human experts, with improved detection of subtle evidentiary nuances. To our knowledge, FEAT is the first LLM-based AI agent system dedicated to forensic medicine, offering scalable, consistent death certification while maintaining expert-level rigor. By integrating AI efficiency with human oversight, this work could advance equitable access to reliable medicolegal services while addressing critical capacity constraints in forensic systems.

Zhehao Huang, Yuhang Liu, Yixin Lou et al.

Continual post-training adapts a single text-to-image diffusion model to learn new tasks without incurring the cost of separate models, but naive post-training causes forgetting of pretrained knowledge and undermines zero-shot compositionality. We observe that the absence of a standardized evaluation protocol hampers related research for continual post-training. To address this, we introduce T2I-ConBench, a unified benchmark for continual post-training of text-to-image models. T2I-ConBench focuses on two practical scenarios, item customization and domain enhancement, and analyzes four dimensions: (1) retention of generality, (2) target-task performance, (3) catastrophic forgetting, and (4) cross-task generalization. It combines automated metrics, human-preference modeling, and vision-language QA for comprehensive assessment. We benchmark ten representative methods across three realistic task sequences and find that no approach excels on all fronts. Even joint "oracle" training does not succeed for every task, and cross-task generalization remains unsolved. We release all datasets, code, and evaluation tools to accelerate research in continual post-training for text-to-image models.

Qiaowei Miao, Kehan Li, Jinsheng Quan et al.

Generative artificial intelligence has recently progressed from static image and video synthesis to 3D content generation, culminating in the emergence of 4D generation-the task of synthesizing temporally coherent dynamic 3D assets guided by user input. As a burgeoning research frontier, 4D generation enables richer interactive and immersive experiences, with applications ranging from digital humans to autonomous driving. Despite rapid progress, the field lacks a unified understanding of 4D representations, generative frameworks, basic paradigms, and the core technical challenges it faces. This survey provides a systematic and in-depth review of the 4D generation landscape. To comprehensively characterize 4D generation, we first categorize fundamental 4D representations and outline associated techniques for 4D generation. We then present an in-depth analysis of representative generative pipelines based on conditions and representation methods. Subsequently, we discuss how motion and geometry priors are integrated into 4D outputs to ensure spatio-temporal consistency under various control schemes. From an application perspective, this paper summarizes 4D generation tasks in areas such as dynamic object/scene generation, digital human synthesis, editable 4D content, and embodied AI. Furthermore, we summarize and multi-dimensionally compare four basic paradigms for 4D generation: End-to-End, Generated-Data-Based, Implicit-Distillation-Based, and Explicit-Supervision-Based. Concluding our analysis, we highlight five key challenges-consistency, controllability, diversity, efficiency, and fidelity-and contextualize these with current approaches.By distilling recent advances and outlining open problems, this work offers a comprehensive and forward-looking perspective to guide future research in 4D generation.

Zesen Cheng, Kehan Li, Hao Li et al.

Temporally locating objects with arbitrary class texts is the primary pursuit of open-vocabulary Video Instance Segmentation (VIS). Because of the insufficient vocabulary of video data, previous methods leverage image-text pretraining model for recognizing object instances by separately aligning each frame and class texts, ignoring the correlation between frames. As a result, the separation breaks the instance movement context of videos, causing inferior alignment between video and text. To tackle this issue, we propose to link frame-level instance representations as a Brownian Bridge to model instance dynamics and align bridge-level instance representation to class texts for more precisely open-vocabulary VIS (BriVIS). Specifically, we build our system upon a frozen video segmentor to generate frame-level instance queries, and design Temporal Instance Resampler (TIR) to generate queries with temporal context from frame queries. To mold instance queries to follow Brownian bridge and accomplish alignment with class texts, we design Bridge-Text Alignment (BTA) to learn discriminative bridge-level representations of instances via contrastive objectives. Setting MinVIS as the basic video segmentor, BriVIS surpasses the Open-vocabulary SOTA (OV2Seg) by a clear margin. For example, on the challenging large-vocabulary VIS dataset (BURST), BriVIS achieves 7.43 mAP and exhibits 49.49% improvement compared to OV2Seg (4.97 mAP).