Zhiyuan Zhao

Pan Zhang, Xiaoyi Dong, Bin Wang et al. · pku

We propose InternLM-XComposer, a vision-language large model that enables advanced image-text comprehension and composition. The innovative nature of our model is highlighted by three appealing properties: 1) Interleaved Text-Image Composition: InternLM-XComposer can effortlessly generate coherent and contextual articles that seamlessly integrate images, providing a more engaging and immersive reading experience. Simply provide a writing instruction, and our system will generate the corresponding manuscript. It can intelligently identify the areas in the text where images would enhance the content and automatically insert the most appropriate visual candidates. 2) Comprehension with Rich Multilingual Knowledge: The text-image comprehension is empowered by training on an extensive multi-modal multilingual database with carefully crafted strategies, resulting in a deep understanding of visual content. 3) State-of-the-art Performance: Our model consistently achieves state-of-the-art results across various mainstream benchmarks for vision-language foundational models, including MME Benchmark, MMBench, MMBench-CN, Seed-Bench, CCBench (Chinese Cultural Benchmark), QBench and Tiny LVLM. Owing to the absence of established metrics for quantitatively assessing text-image composition, we have devised a robust evaluation procedure that comprises both human and GPT4-Vision (GPT4-V) to ensure reliability. Notably, our InternLM-XComposer achieves competitive text-image composition scores compared to public solutions, including GPT4-V and GPT3.5. Collectively, InternLM-XComposer seamlessly blends advanced text-image comprehension and composition, revolutionizing vision-language interaction and offering new insights and opportunities. The InternLM-XComposer model series are publicly available at https://github.com/InternLM/InternLM-XComposer.

Bin Wang, Chao Xu, Xiaomeng Zhao et al.

Document content analysis has been a crucial research area in computer vision. Despite significant advancements in methods such as OCR, layout detection, and formula recognition, existing open-source solutions struggle to consistently deliver high-quality content extraction due to the diversity in document types and content. To address these challenges, we present MinerU, an open-source solution for high-precision document content extraction. MinerU leverages the sophisticated PDF-Extract-Kit models to extract content from diverse documents effectively and employs finely-tuned preprocessing and postprocessing rules to ensure the accuracy of the final results. Experimental results demonstrate that MinerU consistently achieves high performance across various document types, significantly enhancing the quality and consistency of content extraction. The MinerU open-source project is available at https://github.com/opendatalab/MinerU.

Zhiyuan Zhao, Qingjie Liu, Yunhong Wang

Recently, few-shot object detection~(FSOD) has received much attention from the community, and many methods are proposed to address this problem from a knowledge transfer perspective. Though promising results have been achieved, these methods fail to achieve shot-stable:~methods that excel in low-shot regimes are likely to struggle in high-shot regimes, and vice versa. We believe this is because the primary challenge of FSOD changes when the number of shots varies. In the low-shot regime, the primary challenge is the lack of inner-class variation. In the high-shot regime, as the variance approaches the real one, the main hindrance to the performance comes from misalignment between learned and true distributions. However, these two distinct issues remain unsolved in most existing FSOD methods. In this paper, we propose to overcome these challenges by exploiting rich knowledge the model has learned and effectively transferring them to the novel classes. For the low-shot regime, we propose a distribution calibration method to deal with the lack of inner-class variation problem. Meanwhile, a shift compensation method is proposed to compensate for possible distribution shift during fine-tuning. For the high-shot regime, we propose to use the knowledge learned from ImageNet as guidance for the feature learning in the fine-tuning stage, which will implicitly align the distributions of the novel classes. Although targeted toward different regimes, these two strategies can work together to further improve the FSOD performance. Experiments on both the VOC and COCO benchmarks show that our proposed method can significantly outperform the baseline method and produce competitive results in both low-shot settings (shot<5) and high-shot settings (shot>=5). Code is available at https://github.com/JulioZhao97/EffTrans_Fsdet.git.

Yicheng Zou, Dongsheng Zhu, Lin Zhu et al.

We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, including chemistry, materials, life sciences, and earth sciences. Achieving this massive scale is made possible by the robust infrastructure support of XTuner and LMDeploy, which facilitates highly efficient Reinforcement Learning (RL) training at the 1-trillion parameter level while ensuring strict precision consistency between training and inference. By seamlessly integrating these advancements, Intern-S1-Pro further fortifies the fusion of general and specialized intelligence, working as a Specializable Generalist, demonstrating its position in the top tier of open-source models for general capabilities, while outperforming proprietary models in the depth of specialized scientific tasks.

Zhiyuan Zhao, Linke Ouyang, Bin Wang et al.

Despite the great advance of Multimodal Large Language Models (MLLMs) in both instruction dataset building and benchmarking, the independence of training and evaluation makes current MLLMs hard to further improve their capability under the guidance of evaluation results with a relatively low human cost. In this paper, we propose MLLM-DataEngine, a novel closed-loop system that bridges data generation, model training, and evaluation. Within each loop iteration, the MLLM-DataEngine first analyze the weakness of the model based on the evaluation results, then generate a proper incremental dataset for the next training iteration and enhance the model capability iteratively. Compared with previous data collection methods which are separate from the benchmarking, the data generated by MLLM-DataEngine shows better targeting, quality, and correctness. For targeting, we propose an Adaptive Bad-case Sampling module, which adjusts the ratio of different types of data within each incremental dataset based on the benchmarking results. For quality, we resort to GPT-4 to generate high-quality data with each given data type. For correctness, prompt design is critical for the data generation results. Rather than previous hand-crafted prompt, we propose an Interactive Prompt Optimization strategy, which optimizes the prompt with the multi-round interaction between human and GPT, and improve the correctness of generated data greatly. Through extensive experiments, we find our MLLM-DataEngine could boost the MLLM capability in a targeted and automatic manner, with only a few human participation. We hope it could be a general solution for the following MLLMs building. The MLLM-DataEngine has been open-sourced and is now available at https://github.com/opendatalab/MLLM-DataEngine.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Multimodal semantic segmentation shows significant potential for enhancing segmentation accuracy in complex scenes. However, current methods often incorporate specialized feature fusion modules tailored to specific modalities, thereby restricting input flexibility and increasing the number of training parameters. To address these challenges, we propose StitchFusion, a straightforward yet effective modal fusion framework that integrates large-scale pre-trained models directly as encoders and feature fusers. This approach facilitates comprehensive multi-modal and multi-scale feature fusion, accommodating any visual modal inputs. Specifically, Our framework achieves modal integration during encoding by sharing multi-modal visual information. To enhance information exchange across modalities, we introduce a multi-directional adapter module (MultiAdapter) to enable cross-modal information transfer during encoding. By leveraging MultiAdapter to propagate multi-scale information across pre-trained encoders during the encoding process, StitchFusion achieves multi-modal visual information integration during encoding. Extensive comparative experiments demonstrate that our model achieves state-of-the-art performance on four multi-modal segmentation datasets with minimal additional parameters. Furthermore, the experimental integration of MultiAdapter with existing Feature Fusion Modules (FFMs) highlights their complementary nature. Our code is available at StitchFusion_repo.

Zhiyuan Zhao, Bin Wang, Linke Ouyang et al.

Multimodal large language models have made significant advancements in recent years, yet they still suffer from a common issue known as the "hallucination problem", in which the models generate textual descriptions that inaccurately depict or entirely fabricate content from associated images. This paper introduces a novel solution, Hallucination-Aware Direct Preference Optimization (HA-DPO), which reframes the hallucination problem as a preference selection task. The model is trained to favor the non-hallucinating response when presented with two responses of the same image (one accurate and one hallucinatory). Furthermore, this paper proposes an efficient pipeline for constructing positive~(non-hallucinatory) and negative~(hallucinatory) sample pairs, ensuring a high-quality, style-consistent dataset for robust preference learning. When applied to three mainstream multimodal models, HA-DPO significantly reduced hallucination issues and amplified the models' generalization capabilities. Notably, the MiniGPT-4 model, when enhanced with HA-DPO, demonstrated a substantial improvement: POPE accuracy rose from 51.13% to 86.13% (an absolute improvement of 35%), and the MME score surged from 932.00 to 1326.46 (a relative improvement of 42.32%). The codes, models, and datasets are made accessible at https://opendatalab.github.io/HA-DPO.

Da Zhang, Bingyu Li, Zhiyuan Zhao et al.

Medical time series are central to healthcare, enabling continuous monitoring and supporting timely clinical decisions. Despite recent progress, existing methods struggle to jointly model local-global dynamics and handle nonstationarities like baseline drift, while often failing to capture latent channel interactions. To address these challenges, we propose MedMamba, an end-to-end architecture that integrates state space models with domain-specific inductive biases. Specifically, MedMamba first employs multi-scale convolutional embeddings to capture discriminative local morphology. Second, to mitigate nonstationarity, we introduce a tri-branch differential state space encoder that processes raw, temporal-difference, and frequency-domain views, fusing them to emphasize informative patterns while suppressing drift. Furthermore, to uncover latent channel correlations, we design a spatial graph Mamba module that learns a directed dependency structure regularized toward sparsity and acyclicity, which obviates the need for predefined graphs. Extensive experiments on five real-world datasets demonstrate that MedMamba achieves state-of-the-art performance while maintaining linear computational complexity, and ablation studies validate each component's contribution.Code is available at https://github.com/zhangda1018/MedMamba.

Hengrui Kang, Zhuangcheng Gu, Zhiyuan Zhao et al.

Document AI has advanced rapidly and is attracting increasing attention. Yet, while most efforts have focused on document layout analysis (DLA), its generative counterpart, document layout generation, remains underexplored. A major obstacle lies in the scarcity of diverse layouts: academic papers with Manhattan-style structures dominate existing studies, while open-world genres such as newspapers and magazines remain severely underrepresented. To address this gap, we curate OmniLayout-1M, the first million-scale dataset of diverse document layouts, covering six common document types and comprising contemporary layouts collected from multiple sources. Moreover, since existing methods struggle in complex domains and often fail to arrange long sequences coherently, we introduce OmniLayout-LLM, a 0.5B model with designed two-stage Coarse-to-Fine learning paradigm: 1) learning universal layout principles from OmniLayout-1M with coarse category definitions, and 2) transferring the knowledge to a specific domain with fine-grained annotations. Extensive experiments demonstrate that our approach achieves strong performance on multiple domains in M$^{6}$Doc dataset, substantially surpassing both existing layout generation experts and several latest general-purpose LLMs. Our code, models, and dataset will be publicly released.

Zhiyuan Zhao, Xueying Ding, B. Aditya Prakash

Physics-Informed Neural Networks (PINNs) have emerged as a promising deep learning framework for approximating numerical solutions to partial differential equations (PDEs). However, conventional PINNs, relying on multilayer perceptrons (MLP), neglect the crucial temporal dependencies inherent in practical physics systems and thus fail to propagate the initial condition constraints globally and accurately capture the true solutions under various scenarios. In this paper, we introduce a novel Transformer-based framework, termed PINNsFormer, designed to address this limitation. PINNsFormer can accurately approximate PDE solutions by utilizing multi-head attention mechanisms to capture temporal dependencies. PINNsFormer transforms point-wise inputs into pseudo sequences and replaces point-wise PINNs loss with a sequential loss. Additionally, it incorporates a novel activation function, Wavelet, which anticipates Fourier decomposition through deep neural networks. Empirical results demonstrate that PINNsFormer achieves superior generalization ability and accuracy across various scenarios, including PINNs failure modes and high-dimensional PDEs. Moreover, PINNsFormer offers flexibility in integrating existing learning schemes for PINNs, further enhancing its performance.

Wenming Weng, Ruoyu Feng, Yanhui Wang et al.

We present ART$\boldsymbol{\cdot}$V, an efficient framework for auto-regressive video generation with diffusion models. Unlike existing methods that generate entire videos in one-shot, ART$\boldsymbol{\cdot}$V generates a single frame at a time, conditioned on the previous ones. The framework offers three distinct advantages. First, it only learns simple continual motions between adjacent frames, therefore avoiding modeling complex long-range motions that require huge training data. Second, it preserves the high-fidelity generation ability of the pre-trained image diffusion models by making only minimal network modifications. Third, it can generate arbitrarily long videos conditioned on a variety of prompts such as text, image or their combinations, making it highly versatile and flexible. To combat the common drifting issue in AR models, we propose masked diffusion model which implicitly learns which information can be drawn from reference images rather than network predictions, in order to reduce the risk of generating inconsistent appearances that cause drifting. Moreover, we further enhance generation coherence by conditioning it on the initial frame, which typically contains minimal noise. This is particularly useful for long video generation. When trained for only two weeks on four GPUs, ART$\boldsymbol{\cdot}$V already can generate videos with natural motions, rich details and a high level of aesthetic quality. Besides, it enables various appealing applications, e.g., composing a long video from multiple text prompts.

Bingyu Li, Da Zhang, Tao Huo et al.

Large Vision-Language Models (LVLMs) have shown strong visual understanding and language-guided grounding abilities, yet their capacity for multi-temporal visual reasoning remains underexplored. To bridge this gap, we introduce \textbf{Multi-temporal Referring Segmentation (MTRS)}, a new task that aims to segment language-described temporal changes from multi-temporal images. MTRS extends conventional referring segmentation and change detection by jointly requiring temporal correspondence reasoning, language grounding, and pixel-level mask prediction. We propose \textbf{CRAFT-Agent}, an automated data construction pipeline with human auditing, and build \textbf{MTRefSeg-21K}, the first MTRS benchmark, containing 21K high-quality multi-temporal image-text-mask triplets across diverse scenes, viewpoints, and domains. Benchmarking a broad set of VLM- and LVLM-based models reveals that direct inference performs poorly, while task-specific fine-tuning remains limited. To address this, we propose \textbf{MTRefSeg-R1}, a change-aware LVLM framework trained with a two-stage strategy. It first learns general temporal-change perception from 20K vision-only bi-temporal samples, and is then fine-tuned on MTRefSeg-21K for fine-grained language-guided temporal localization. MTRefSeg-R1 explicitly models cross-temporal visual differences, aligns language instructions with temporal variations, and predicts referred change masks. Extensive experiments show that MTRefSeg-R1 achieves strong and often superior performance compared with existing LVLM baselines, demonstrating the challenge and potential of MTRS.

Bingyu Li, Tao Huo, Haocheng Dong et al.

Open-vocabulary remote sensing image segmentation (OVRSIS) remains underexplored due to fragmented datasets, limited training diversity, and the lack of evaluation benchmarks that reflect realistic geospatial application demands. Our previous \textit{OVRSISBenchV1} established an initial cross-dataset evaluation protocol, but its limited scope is insufficient for assessing realistic open-world generalization. To address this issue, we propose \textit{OVRSISBenchV2}, a large-scale and application-oriented benchmark for OVRSIS. We first construct \textbf{OVRSIS95K}, a balanced dataset of about 95K image--mask pairs covering 35 common semantic categories across diverse remote sensing scenes. Built upon OVRSIS95K and 10 downstream datasets, OVRSISBenchV2 contains 170K images and 128 categories, substantially expanding scene diversity, semantic coverage, and evaluation difficulty. Beyond standard open-vocabulary segmentation, it further includes downstream protocols for building extraction, road extraction, and flood detection, thereby better reflecting realistic geospatial application demands and complex deployment scenarios. We also propose \textbf{Pi-Seg}, a baseline for OVRSIS. Pi-Seg improves transferability through a \textbf{positive-incentive noise} mechanism, where learnable and semantically guided perturbations broaden the visual-text feature space during training. Extensive experiments on OVRSISBenchV1, OVRSISBenchV2, and downstream tasks show that Pi-Seg delivers strong and consistent results, particularly on the more challenging OVRSISBenchV2 benchmark. Our results highlight both the importance of realistic benchmark design and the effectiveness of perturbation-based transfer for OVRSIS. The code and datasets are available at \href{https://github.com/LiBingyu01/RSKT-Seg/tree/Pi-Seg}{LiBingyu01/RSKT-Seg/tree/Pi-Seg}.

Zichun Xu, Zhaomin Wang, Yuntao Li et al.

For peg-in-hole tasks, humans rely on binocular visual perception to locate the peg above the hole surface and then proceed with insertion. This paper draws insights from this behavior to enable agents to learn efficient assembly strategies through visual reinforcement learning. Hence, we propose a Separate Primitive Policy (S2P) to learn how to derive location and insertion actions simultaneously. S2P is compatible with model-free reinforcement learning algorithms. Ten insertion tasks featuring different polygons are developed as benchmarks for evaluations. Simulation experiments show that S2P can boost the sample efficiency and success rate even with force constraints. Real-world experiments are also performed to verify the feasibility of S2P. Ablations are finally given to discuss the generalizability of S2P and some factors that affect its performance.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Text has become the predominant form of communication on social media, embedding a wealth of emotional nuances. Consequently, the extraction of emotional information from text is of paramount importance. Despite previous research making some progress, existing text sentiment analysis models still face challenges in integrating diverse semantic information and lack interpretability. To address these issues, we propose a quantum-inspired deep learning architecture that combines fundamental principles of quantum mechanics (QM principles) with deep learning models for text sentiment analysis. Specifically, we analyze the commonalities between text representation and QM principles to design a quantum-inspired text representation method and further develop a quantum-inspired text embedding layer. Additionally, we design a feature extraction layer based on long short-term memory (LSTM) networks and self-attention mechanisms (SAMs). Finally, we calculate the text density matrix using the quantum complex numbers principle and apply 2D-convolution neural networks (CNNs) for feature condensation and dimensionality reduction. Through a series of visualization, comparative, and ablation experiments, we demonstrate that our model not only shows significant advantages in accuracy and efficiency compared to previous related models but also achieves a certain level of interpretability by integrating QM principles. Our code is available at QISA.

Zhiyuan Zhao, Juntong Ni, Shangqing Xu et al.

Time-series forecasting is an essential task with wide real-world applications across domains. While recent advances in deep learning have enabled time-series forecasting models with accurate predictions, there remains considerable debate over which architectures and design components, such as series decomposition or normalization, are most effective under varying conditions. Existing benchmarks primarily evaluate models at a high level, offering limited insight into why certain designs work better. To mitigate this gap, we propose TimeRecipe, a unified benchmarking framework that systematically evaluates time-series forecasting methods at the module level. TimeRecipe conducts over 10,000 experiments to assess the effectiveness of individual components across a diverse range of datasets, forecasting horizons, and task settings. Our results reveal that exhaustive exploration of the design space can yield models that outperform existing state-of-the-art methods and uncover meaningful intuitions linking specific design choices to forecasting scenarios. Furthermore, we release a practical toolkit within TimeRecipe that recommends suitable model architectures based on these empirical insights. The benchmark is available at: https://github.com/AdityaLab/TimeRecipe.

Da Zhang, Bingyu Li, Zhiyuan Zhao et al.

Time series analysis plays a vital role in fields such as finance, healthcare, industry, and meteorology, underpinning key tasks including classification, forecasting, and anomaly detection. Although deep learning models have achieved remarkable progress in these areas in recent years, constructing an efficient, multi-task compatible, and generalizable unified framework for time series analysis remains a significant challenge. Existing approaches are often tailored to single tasks or specific data types, making it difficult to simultaneously handle multi-task modeling and effectively integrate information across diverse time series types. Moreover, real-world data are often affected by noise, complex frequency components, and multi-scale dynamic patterns, which further complicate robust feature extraction and analysis. To ameliorate these challenges, we propose FusAD, a unified analysis framework designed for diverse time series tasks. FusAD features an adaptive time-frequency fusion mechanism, integrating both Fourier and Wavelet transforms to efficiently capture global-local and multi-scale dynamic features. With an adaptive denoising mechanism, FusAD automatically senses and filters various types of noise, highlighting crucial sequence variations and enabling robust feature extraction in complex environments. In addition, the framework integrates a general information fusion and decoding structure, combined with masked pre-training, to promote efficient learning and transfer of multi-granularity representations. Extensive experiments demonstrate that FusAD consistently outperforms state-of-the-art models on mainstream time series benchmarks for classification, forecasting, and anomaly detection tasks, while maintaining high efficiency and scalability. Code is available at https://github.com/zhangda1018/FusAD.

Shangqing Xu, Zhiyuan Zhao, Megha Sharma et al.

Severe heatwaves in urban areas significantly threaten public health, calling for establishing early warning strategies. Despite predicting occurrence of heatwaves and attributing historical mortality, predicting an incoming deadly heatwave remains a challenge due to the difficulty in defining and estimating heat-related mortality. Furthermore, establishing an early warning system imposes additional requirements, including data availability, spatial and temporal robustness, and decision costs. To address these challenges, we propose DeepTherm, a modular early warning system for deadly heatwave prediction without requiring heat-related mortality history. By highlighting the flexibility of deep learning, DeepTherm employs a dual-prediction pipeline, disentangling baseline mortality in the absence of heatwaves and other irregular events from all-cause mortality. We evaluated DeepTherm on real-world data across Spain. Results demonstrate consistent, robust, and accurate performance across diverse regions, time periods, and population groups while allowing trade-off between missed alarms and false alarms.

Wenxun Dai, Zhiyuan Zhao, Yule Zhong et al.

Unified multimodal models (UMMs) have achieved remarkable progress yet remain constrained by a single-turn interaction paradigm, effectively functioning as solvers for independent requests rather than assistants in continuous dialogue. To bridge this gap, we present ChatUMM. As a conversational unified model, it excels at robust context tracking to sustain interleaved multimodal generation. ChatUMM derives its capabilities from two key innovations: an interleaved multi-turn training strategy that models serialized text-image streams as a continuous conversational flow, and a systematic conversational data synthesis pipeline. This pipeline transforms a diverse set of standard single-turn datasets into fluid dialogues through three progressive stages: constructing basic stateful dialogues, enforcing long-range dependency resolution via ``distractor'' turns with history-dependent query rewriting, and synthesizing naturally interleaved multimodal responses. Extensive evaluations demonstrate that ChatUMM achieves state-of-the-art performance among open-source unified models on visual understanding and instruction-guided editing benchmarks, while maintaining competitive fidelity in text-to-image generation. Notably, ChatUMM exhibits superior robustness in complex multi-turn scenarios, ensuring fluid, context-aware dialogues.

Zhiyuan Zhao, Hengrui Kang, Bin Wang et al.

Document Layout Analysis is crucial for real-world document understanding systems, but it encounters a challenging trade-off between speed and accuracy: multimodal methods leveraging both text and visual features achieve higher accuracy but suffer from significant latency, whereas unimodal methods relying solely on visual features offer faster processing speeds at the expense of accuracy. To address this dilemma, we introduce DocLayout-YOLO, a novel approach that enhances accuracy while maintaining speed advantages through document-specific optimizations in both pre-training and model design. For robust document pre-training, we introduce the Mesh-candidate BestFit algorithm, which frames document synthesis as a two-dimensional bin packing problem, generating the large-scale, diverse DocSynth-300K dataset. Pre-training on the resulting DocSynth-300K dataset significantly improves fine-tuning performance across various document types. In terms of model optimization, we propose a Global-to-Local Controllable Receptive Module that is capable of better handling multi-scale variations of document elements. Furthermore, to validate performance across different document types, we introduce a complex and challenging benchmark named DocStructBench. Extensive experiments on downstream datasets demonstrate that DocLayout-YOLO excels in both speed and accuracy. Code, data, and models are available at https://github.com/opendatalab/DocLayout-YOLO.

Zhiyuan Zhao, Haoxin Liu, Alexander Rodriguez et al.

Time-series forecasting is a critical challenge in various domains and has witnessed substantial progress in recent years. Many real-life scenarios, such as public health, economics, and social applications, involve feedback loops where predictions can influence the predicted outcome, subsequently altering the target variable's distribution. This phenomenon, known as performativity, introduces the potential for 'self-negating' or 'self-fulfilling' predictions. Despite extensive studies in classification problems across domains, performativity remains largely unexplored in the context of time-series forecasting from a machine-learning perspective. In this paper, we formalize performative time-series forecasting (PeTS), addressing the challenge of accurate predictions when performativity-induced distribution shifts are possible. We propose a novel approach, Feature Performative-Shifting (FPS), which leverages the concept of delayed response to anticipate distribution shifts and subsequently predicts targets accordingly. We provide theoretical insights suggesting that FPS can potentially lead to reduced generalization error. We conduct comprehensive experiments using multiple time-series models on COVID-19 and traffic forecasting tasks. The results demonstrate that FPS consistently outperforms conventional time-series forecasting methods, highlighting its efficacy in handling performativity-induced challenges.

Linke Ouyang, Yuan Qu, Hongbin Zhou et al.

Document content extraction is a critical task in computer vision, underpinning the data needs of large language models (LLMs) and retrieval-augmented generation (RAG) systems. Despite recent progress, current document parsing methods have not been fairly and comprehensively evaluated due to the narrow coverage of document types and the simplified, unrealistic evaluation procedures in existing benchmarks. To address these gaps, we introduce OmniDocBench, a novel benchmark featuring high-quality annotations across nine document sources, including academic papers, textbooks, and more challenging cases such as handwritten notes and densely typeset newspapers. OmniDocBench supports flexible, multi-level evaluations--ranging from an end-to-end assessment to the task-specific and attribute--based analysis using 19 layout categories and 15 attribute labels. We conduct a thorough evaluation of both pipeline-based methods and end-to-end vision-language models, revealing their strengths and weaknesses across different document types. OmniDocBench sets a new standard for the fair, diverse, and fine-grained evaluation in document parsing. Dataset and code are available at https://github.com/opendatalab/OmniDocBench.

Haoxin Liu, Harshavardhan Kamarthi, Zhiyuan Zhao et al.

Shot language understanding (SLU) is crucial for cinematic analysis but remains challenging due to its diverse cinematographic dimensions and subjective expert judgment. While vision-language models (VLMs) have shown strong ability in general visual understanding, recent studies reveal judgment discrepancies between VLMs and film experts on SLU tasks. To address this gap, we introduce SLU-SUITE, a comprehensive training and evaluation suite containing 490K human-annotated QA pairs across 33 tasks spanning six film-grounded dimensions. Using SLU-SUITE, we originally observe two insights into VLM-based SLU from: the model side, which diagnoses key bottlenecks of modules; the data side, which quantifies cross-dimensional influences among tasks. These findings motivate our universal SLU solutions from two complementary paradigms: UniShot, a balanced one-for-all generalist trained via dynamic-balanced data mixing, and AgentShots, a prompt-routed expert cluster that maximizes peak dimension performance. Extensive experiments show that our models outperform task-specific ensembles on in-domain tasks and surpass leading commercial VLMs by 22% on out-of-domain tasks.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Multimodal semantic segmentation is a pivotal component of computer vision and typically surpasses unimodal methods by utilizing rich information set from various sources.Current models frequently adopt modality-specific frameworks that inherently biases toward certain modalities. Although these biases might be advantageous in specific situations, they generally limit the adaptability of the models across different multimodal contexts, thereby potentially impairing performance. To address this issue, we leverage the inherent capabilities of the model itself to discover the optimal equilibrium in multimodal fusion and introduce U3M: An Unbiased Multiscale Modal Fusion Model for Multimodal Semantic Segmentation. Specifically, this method involves an unbiased integration of multimodal visual data. Additionally, we employ feature fusion at multiple scales to ensure the effective extraction and integration of both global and local features. Experimental results demonstrate that our approach achieves superior performance across multiple datasets, verifing its efficacy in enhancing the robustness and versatility of semantic segmentation in diverse settings. Our code is available at U3M-multimodal-semantic-segmentation.

Haoxin Liu, Zhiyuan Zhao, Shiduo Li et al. · gatech

Reasoning ability is crucial for solving challenging tasks. With the advancement of foundation models, such as the emergence of large language models (LLMs), a wide range of reasoning strategies has been proposed, including test-time enhancements, such as Chain-ofThought, and post-training optimizations, as used in DeepSeek-R1. While these reasoning strategies have demonstrated effectiveness across various challenging language or vision tasks, their applicability and impact on time-series forecasting (TSF), particularly the challenging zero-shot TSF, remain largely unexplored. In particular, it is unclear whether zero-shot TSF benefits from reasoning and, if so, what types of reasoning strategies are most effective. To bridge this gap, we propose ReC4TS, the first benchmark that systematically evaluates the effectiveness of popular reasoning strategies when applied to zero-shot TSF tasks. ReC4TS conducts comprehensive evaluations across datasets spanning eight domains, covering both unimodal and multimodal with short-term and longterm forecasting tasks. More importantly, ReC4TS provides key insights: (1) Self-consistency emerges as the most effective test-time reasoning strategy; (2) Group-relative policy optimization emerges as a more suitable approach for incentivizing reasoning ability during post-training; (3) Multimodal TSF benefits more from reasoning strategies compared to unimodal TSF. Beyond these insights, ReC4TS establishes two pioneering starting blocks to support future zero-shot TSF reasoning research: (1) A novel dataset, TimeThinking, containing forecasting samples annotated with reasoning trajectories from multiple advanced LLMs, and (2) A new and simple test-time scaling-law validated on foundational TSF models enabled by self-consistency reasoning strategy. All data and code are publicly accessible at: https://github.com/AdityaLab/OpenTimeR

Da Zhang, Bingyu Li, Feiyu Wang et al.

Zero-shot object counting (ZSOC) aims to enumerate objects of arbitrary categories specified by text descriptions without requiring visual exemplars. However, existing methods often treat counting as a coarse retrieval task, suffering from a lack of fine-grained quantity awareness. Furthermore, they frequently exhibit spatial insensitivity and degraded generalization due to feature space distortion during model adaptation.To address these challenges, we present \textbf{QICA}, a novel framework that synergizes \underline{q}uantity percept\underline{i}on with robust spatial \underline{c}ast \underline{a}ggregation. Specifically, we introduce a Synergistic Prompting Strategy (\textbf{SPS}) that adapts vision and language encoders through numerically conditioned prompts, bridging the gap between semantic recognition and quantitative reasoning. To mitigate feature distortion, we propose a Cost Aggregation Decoder (\textbf{CAD}) that operates directly on vision-text similarity maps. By refining these maps through spatial aggregation, CAD prevents overfitting while preserving zero-shot transferability. Additionally, a multi-level quantity alignment loss ($\mathcal{L}_{MQA}$) is employed to enforce numerical consistency across the entire pipeline. Extensive experiments on FSC-147 demonstrate competitive performance, while zero-shot evaluation on CARPK and ShanghaiTech-A validates superior generalization to unseen domains.

Siqu Ou, Tianrui Wan, Zhiyuan Zhao et al.

While chain-of-thought (CoT) reasoning has substantially improved multimodal large language models (MLLMs) on complex reasoning tasks, existing approaches largely rely on long textual reasoning trajectories and provide limited mechanisms for learning stable visual attention policies. Our analysis shows that current MLLMs exhibit weak visual focus: early-stage visual misalignment is rarely corrected during subsequent reasoning, leading to error propagation and failed inferences. We argue that this limitation stems from inadequate credit assignment for visual attention during training. To address this issue, we propose SAYO, a visual reasoning model trained with a reinforcement learning (RL) framework that introduces a region-level visual attention-based reward. This reward explicitly aligns optimization signals with visually grounded reasoning steps, enabling the model to learn more reliable attention behaviors. Extensive experiments across multiple multimodal benchmarks demonstrate that SAYO consistently improves performance on diverse reasoning and perception tasks.

Bingyu Li, Tao Huo, Da Zhang et al.

Accurate segmentation of marine organisms is vital for biodiversity monitoring and ecological assessment, yet existing datasets and models remain largely limited to terrestrial scenes. To bridge this gap, we introduce \textbf{AquaOV255}, the first large-scale and fine-grained underwater segmentation dataset containing 255 categories and over 20K images, covering diverse categories for open-vocabulary (OV) evaluation. Furthermore, we establish the first underwater OV segmentation benchmark, \textbf{UOVSBench}, by integrating AquaOV255 with five additional underwater datasets to enable comprehensive evaluation. Alongside, we present \textbf{Earth2Ocean}, a training-free OV segmentation framework that transfers terrestrial vision--language models (VLMs) to underwater domains without any additional underwater training. Earth2Ocean consists of two core components: a Geometric-guided Visual Mask Generator (\textbf{GMG}) that refines visual features via self-similarity geometric priors for local structure perception, and a Category-visual Semantic Alignment (\textbf{CSA}) module that enhances text embeddings through multimodal large language model reasoning and scene-aware template construction. Extensive experiments on the UOVSBench benchmark demonstrate that Earth2Ocean achieves significant performance improvement on average while maintaining efficient inference.

Feiyu Wang, Jiayuan Yang, Zhiyuan Zhao et al.

Scalable Vector Graphics (SVG) are central to digital design due to their inherent scalability and editability. Despite significant advancements in content generation enabled by Visual Language Models (VLMs), existing text-to-SVG generation methods are limited by a core challenge: the autoregressive training process does not incorporate visual perception of the final rendered image, which fundamentally constrains generation quality. To address this limitation, we propose an Introspective SVG Generation Framework (IntroSVG). At its core, the framework instantiates a unified VLM that operates in a closed loop, assuming dual roles of both generator and critic. Specifically, through Supervised Fine-Tuning (SFT), the model learns to draft SVGs and to provide feedback on their rendered outputs; moreover, we systematically convert early-stage failures into high-quality error-correction training data, thereby enhancing model robustness. Subsequently, we leverage a high-capacity teacher VLM to construct a preference dataset and further align the generator's policy through Direct Preference Optimization (DPO). During inference, the optimized generator and critic operate collaboratively in an iterative "generate-review-refine" cycle, starting from imperfect intermediate drafts to autonomously improve output quality. Experimental results demonstrate that our method achieves state-of-the-art performance across several key evaluation metrics, generating SVGs with more complex structures, stronger semantic alignment, and greater editability. These results corroborate the effectiveness of incorporating explicit visual feedback into the generation loop.

Siyu Cao, Hangting Chen, Peng Chen et al.

We present HunyuanImage 3.0, a native multimodal model that unifies multimodal understanding and generation within an autoregressive framework, with its image generation module publicly available. The achievement of HunyuanImage 3.0 relies on several key components, including meticulous data curation, advanced architecture design, a native Chain-of-Thoughts schema, progressive model pre-training, aggressive model post-training, and an efficient infrastructure that enables large-scale training and inference. With these advancements, we successfully trained a Mixture-of-Experts (MoE) model comprising over 80 billion parameters in total, with 13 billion parameters activated per token during inference, making it the largest and most powerful open-source image generative model to date. We conducted extensive experiments and the results of automatic and human evaluation of text-image alignment and visual quality demonstrate that HunyuanImage 3.0 rivals previous state-of-the-art models. By releasing the code and weights of HunyuanImage 3.0, we aim to enable the community to explore new ideas with a state-of-the-art foundation model, fostering a dynamic and vibrant multimodal ecosystem. All open source assets are publicly available at https://github.com/Tencent-Hunyuan/HunyuanImage-3.0

Bingyu Li, Haocheng Dong, Da Zhang et al.

Open-Vocabulary Remote Sensing Image Segmentation (OVRSIS), an emerging task that adapts Open-Vocabulary Segmentation (OVS) to the remote sensing (RS) domain, remains underexplored due to the absence of a unified evaluation benchmark and the domain gap between natural and RS images. To bridge these gaps, we first establish a standardized OVRSIS benchmark (\textbf{OVRSISBench}) based on widely-used RS segmentation datasets, enabling consistent evaluation across methods. Using this benchmark, we comprehensively evaluate several representative OVS/OVRSIS models and reveal their limitations when directly applied to remote sensing scenarios. Building on these insights, we propose \textbf{RSKT-Seg}, a novel open-vocabulary segmentation framework tailored for remote sensing. RSKT-Seg integrates three key components: (1) a Multi-Directional Cost Map Aggregation (RS-CMA) module that captures rotation-invariant visual cues by computing vision-language cosine similarities across multiple directions; (2) an Efficient Cost Map Fusion (RS-Fusion) transformer, which jointly models spatial and semantic dependencies with a lightweight dimensionality reduction strategy; and (3) a Remote Sensing Knowledge Transfer (RS-Transfer) module that injects pre-trained knowledge and facilitates domain adaptation via enhanced upsampling. Extensive experiments on the benchmark show that RSKT-Seg consistently outperforms strong OVS baselines by +3.8 mIoU and +5.9 mACC, while achieving 2x faster inference through efficient aggregation. Our code is \href{https://github.com/LiBingyu01/RSKT-Seg}{\textcolor{blue}{here}}.

Junyu Gao, Da Zhang, Qiyu Wang et al.

Crowd localization targets on predicting each instance precise location within an image. Current advanced methods propose the pixel-wise binary classification to tackle the congested prediction, in which the pixel-level thresholds binarize the prediction confidence of being the pedestrian head. Since the crowd scenes suffer from extremely varying contents, counts and scales, the confidence-threshold learner is fragile and under-generalized encountering domain knowledge shift. Moreover, at the most time, the target domain is agnostic in training. Hence, it is imperative to exploit how to enhance the generalization of confidence-threshold locator to the latent target domain. In this paper, we propose a Dynamic Proxy Domain (DPD) method to generalize the learner under domain shift. Concretely, based on the theoretical analysis to the generalization error risk upper bound on the latent target domain to a binary classifier, we propose to introduce a generated proxy domain to facilitate generalization. Then, based on the theory, we design a DPD algorithm which is composed by a training paradigm and proxy domain generator to enhance the domain generalization of the confidence-threshold learner. Besides, we conduct our method on five kinds of domain shift scenarios, demonstrating the effectiveness on generalizing the crowd localization. Our code will be available at https://github.com/zhangda1018/DPD.

Da Zhang, Chenggang Rong, Bingyu Li et al.

Large vision-language models (VLMs) have achieved remarkable success in natural scene understanding, yet their application to underwater environments remains largely unexplored. Underwater imagery presents unique challenges including severe light attenuation, color distortion, and suspended particle scattering, while requiring specialized knowledge of marine ecosystems and organism taxonomy. To bridge this gap, we introduce UWBench, a comprehensive benchmark specifically designed for underwater vision-language understanding. UWBench comprises 15,003 high-resolution underwater images captured across diverse aquatic environments, encompassing oceans, coral reefs, and deep-sea habitats. Each image is enriched with human-verified annotations including 15,281 object referring expressions that precisely describe marine organisms and underwater structures, and 124,983 question-answer pairs covering diverse reasoning capabilities from object recognition to ecological relationship understanding. The dataset captures rich variations in visibility, lighting conditions, and water turbidity, providing a realistic testbed for model evaluation. Based on UWBench, we establish three comprehensive benchmarks: detailed image captioning for generating ecologically informed scene descriptions, visual grounding for precise localization of marine organisms, and visual question answering for multimodal reasoning about underwater environments. Extensive experiments on state-of-the-art VLMs demonstrate that underwater understanding remains challenging, with substantial room for improvement. Our benchmark provides essential resources for advancing vision-language research in underwater contexts and supporting applications in marine science, ecological monitoring, and autonomous underwater exploration. Our code and benchmark will be available.

Wenjie Du, Jun Wang, Linglong Qian et al.

Effective imputation is a crucial preprocessing step for time series analysis. Despite the development of numerous deep learning algorithms for time series imputation, the community lacks standardized and comprehensive benchmark platforms to effectively evaluate imputation performance across different settings. Moreover, although many deep learning forecasting algorithms have demonstrated excellent performance, whether their modelling achievements can be transferred to time series imputation tasks remains unexplored. To bridge these gaps, we develop TSI-Bench, the first (to our knowledge) comprehensive benchmark suite for time series imputation utilizing deep learning techniques. The TSI-Bench pipeline standardizes experimental settings to enable fair evaluation of imputation algorithms and identification of meaningful insights into the influence of domain-appropriate missing rates and patterns on model performance. Furthermore, TSI-Bench innovatively provides a systematic paradigm to tailor time series forecasting algorithms for imputation purposes. Our extensive study across 34,804 experiments, 28 algorithms, and 8 datasets with diverse missingness scenarios demonstrates TSI-Bench's effectiveness in diverse downstream tasks and potential to unlock future directions in time series imputation research and analysis. All source code and experiment logs are released at https://github.com/WenjieDu/AwesomeImputation.

Haoxin Liu, Shangqing Xu, Zhiyuan Zhao et al.

Time series data are ubiquitous across a wide range of real-world domains. While real-world time series analysis (TSA) requires human experts to integrate numerical series data with multimodal domain-specific knowledge, most existing TSA models rely solely on numerical data, overlooking the significance of information beyond numerical series. This oversight is due to the untapped potential of textual series data and the absence of a comprehensive, high-quality multimodal dataset. To overcome this obstacle, we introduce Time-MMD, the first multi-domain, multimodal time series dataset covering 9 primary data domains. Time-MMD ensures fine-grained modality alignment, eliminates data contamination, and provides high usability. Additionally, we develop MM-TSFlib, the first-cut multimodal time-series forecasting (TSF) library, seamlessly pipelining multimodal TSF evaluations based on Time-MMD for in-depth analyses. Extensive experiments conducted on Time-MMD through MM-TSFlib demonstrate significant performance enhancements by extending unimodal TSF to multimodality, evidenced by over 15% mean squared error reduction in general, and up to 40% in domains with rich textual data. More importantly, our datasets and library revolutionize broader applications, impacts, research topics to advance TSA. The dataset is available at https://github.com/AdityaLab/Time-MMD.

Haoxin Liu, Zhiyuan Zhao, Jindong Wang et al.

Time-series forecasting (TSF) finds broad applications in real-world scenarios. Prompting off-the-shelf Large Language Models (LLMs) demonstrates strong zero-shot TSF capabilities while preserving computational efficiency. However, existing prompting methods oversimplify TSF as language next-token predictions, overlooking its dynamic nature and lack of integration with state-of-the-art prompt strategies such as Chain-of-Thought. Thus, we propose LSTPrompt, a novel approach for prompting LLMs in zero-shot TSF tasks. LSTPrompt decomposes TSF into short-term and long-term forecasting sub-tasks, tailoring prompts to each. LSTPrompt guides LLMs to regularly reassess forecasting mechanisms to enhance adaptability. Extensive evaluations demonstrate consistently better performance of LSTPrompt than existing prompting methods, and competitive results compared to foundation TSF models.

Bin Wang, Tianyao He, Linke Ouyang et al.

Current document parsing methods compete primarily on model architecture innovation, while systematic engineering of training data remains underexplored. Yet SOTA models of different architectures and parameter scales exhibit highly consistent failure patterns on the same set of hard samples, suggesting that the performance bottleneck stems from shared deficiencies in training data rather than architecture itself. Building on this finding, we present \minerupro, which advances the state of the art solely through data engineering and training strategy optimization while keeping the 1.2B-parameter architecture of \mineru completely fixed. At its core is a Data Engine co-designed around coverage, informativeness, and annotation accuracy: Diversity-and-Difficulty-Aware Sampling expands training data from under 10M to 65.5M samples while correcting distribution shift; Cross-Model Consistency Verification leverages output agreement among heterogeneous models to assess sample difficulty and generate reliable annotations; the Judge-and-Refine pipeline improves annotation quality for hard samples through render-then-verify iterative correction. A three-stage progressive training strategy -- large-scale pre-training, hard sample fine-tuning, and GRPO alignment -- sequentially exploits these data at different quality tiers. On the evaluation front, we fix element-matching biases in OmniDocBench~v1.5 and introduce a Hard subset, establishing the more discriminative OmniDocBench~v1.6 protocol. Without any architectural modification, \minerupro achieves 95.69 on OmniDocBench~v1.6, improving over the same-architecture baseline by 2.71 points and surpassing all existing methods including models with over 200$\times$ more parameters.

Haoxin Liu, Harshavardhan Kamarthi, Zhiyuan Zhao et al.

Time series analysis (TSA) is a longstanding research topic in the data mining community and has wide real-world significance. Compared to "richer" modalities such as language and vision, which have recently experienced explosive development and are densely connected, the time-series modality remains relatively underexplored and isolated. We notice that many recent TSA works have formed a new research field, i.e., Multiple Modalities for TSA (MM4TSA). In general, these MM4TSA works follow a common motivation: how TSA can benefit from multiple modalities. This survey is the first to offer a comprehensive review and a detailed outlook for this emerging field. Specifically, we systematically discuss three benefits: (1) reusing foundation models of other modalities for efficient TSA, (2) multimodal extension for enhanced TSA, and (3) cross-modality interaction for advanced TSA. We further group the works by the introduced modality type, including text, images, audio, tables, and others, within each perspective. Finally, we identify the gaps with future opportunities, including the reused modalities selections, heterogeneous modality combinations, and unseen tasks generalizations, corresponding to the three benefits. We release an up-to-date GitHub repository that includes key papers and resources.

Jiayi Guo, Linqing Wang, Jiangshan Wang et al.

Unified multimodal models (UMMs) integrate visual understanding and generation within a single framework. For text-to-image (T2I) tasks, this unified capability allows UMMs to refine outputs after their initial generation, potentially extending the performance upper bound. Current UMM-based refinement methods primarily follow a refinement-via-editing (RvE) paradigm, where UMMs produce editing instructions to modify misaligned regions while preserving aligned content. However, editing instructions often describe prompt-image misalignment only coarsely, leading to incomplete refinement. Moreover, pixel-level preservation, though necessary for editing, unnecessarily restricts the effective modification space for refinement. To address these limitations, we propose Refinement via Regeneration (RvR), a novel framework that reformulates refinement as conditional image regeneration rather than editing. Instead of relying on editing instructions and enforcing strict content preservation, RvR regenerates images conditioned on the target prompt and the semantic tokens of the initial image, enabling more complete semantic alignment with a larger modification space. Extensive experiments demonstrate the effectiveness of RvR, improving Geneval from 0.78 to 0.91, DPGBench from 84.02 to 87.21, and UniGenBench++ from 61.53 to 77.41.

Muzhi Dai, Jiashuo Sun, Zhiyuan Zhao et al.

Aligning large vision-language models (LVLMs) with human preferences is challenging due to the scarcity of fine-grained, high-quality, and multimodal preference data without human annotations. Existing methods relying on direct distillation often struggle with low-confidence data, leading to suboptimal performance. To address this, we propose CAREVL, a novel method for preference reward modeling by reliably using both high- and low-confidence data. First, a cluster of auxiliary expert models (textual reward models) innovatively leverages image captions as weak supervision signals to filter high-confidence data. The high-confidence data are then used to fine-tune the LVLM. Second, low-confidence data are used to generate diverse preference samples using the fine-tuned LVLM. These samples are then scored and selected to construct reliable chosen-rejected pairs for further training. CAREVL achieves performance improvements over traditional distillation-based methods on VL-RewardBench and MLLM-as-a-Judge benchmark, demonstrating its effectiveness. The code will be released soon.

Muzhi Dai, Shixuan Liu, Zhiyuan Zhao et al.

The rapid advancement of multimodal large language models (MLLMs) has led to breakthroughs in various applications, yet their security remains a critical challenge. One pressing issue involves unsafe image-query pairs--jailbreak inputs specifically designed to bypass security constraints and elicit unintended responses from MLLMs. Compared to general multimodal data, such unsafe inputs are relatively sparse, which limits the diversity and richness of training samples available for developing robust defense models. Meanwhile, existing guardrail-type methods rely on external modules to enforce security constraints but fail to address intrinsic vulnerabilities within MLLMs. Traditional supervised fine-tuning (SFT), on the other hand, often over-refuses harmless inputs, compromising general performance. Given these challenges, we propose Secure Tug-of-War (SecTOW), an innovative iterative defense-attack training method to enhance the security of MLLMs. SecTOW consists of two modules: a defender and an auxiliary attacker, both trained iteratively using reinforcement learning (GRPO). During the iterative process, the attacker identifies security vulnerabilities in the defense model and expands jailbreak data. The expanded data are then used to train the defender, enabling it to address identified security vulnerabilities. We also design reward mechanisms used for GRPO to simplify the use of response labels, reducing dependence on complex generative labels and enabling the efficient use of synthetic data. Additionally, a quality monitoring mechanism is used to mitigate the defender's over-refusal of harmless inputs and ensure the diversity of the jailbreak data generated by the attacker. Experimental results on safety-specific and general benchmarks demonstrate that SecTOW significantly improves security while preserving general performance.

Zhiyu Lin, Zhengda Zhou, Zhiyuan Zhao et al.

With the rapid advancement of Generative AI technology, Multimodal Large Language Models(MLLMs) have the potential to act as AI software engineers capable of executing complex web application development. Considering that the model requires a confluence of multidimensional sub-capabilities to address the challenges of various development phases, constructing a multi-view evaluation framework is crucial for accurately guiding the enhancement of development efficiency. However, existing benchmarks usually fail to provide an assessment of sub-capabilities and focus solely on webpage generation outcomes. In this work, we draw inspiration from the principles of software engineering and further propose WebUIBench, a benchmark systematically designed to evaluate MLLMs in four key areas: WebUI Perception, HTML Programming,WebUI-HTML Understanding, and WebUI-to-Code. WebUIBench comprises 21K high-quality question-answer pairs derived from over 0.7K real-world websites. The extensive evaluation of 29 mainstream MLLMs uncovers the skill characteristics and various weakness that models encountered during the development process.

Junbo Niu, Zheng Liu, Zhuangcheng Gu et al.

We introduce MinerU2.5, a 1.2B-parameter document parsing vision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs a coarse-to-fine, two-stage parsing strategy that decouples global layout analysis from local content recognition. In the first stage, the model performs efficient layout analysis on downsampled images to identify structural elements, circumventing the computational overhead of processing high-resolution inputs. In the second stage, guided by the global layout, it performs targeted content recognition on native-resolution crops extracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensive data engine that generates diverse, large-scale training corpora for both pretraining and fine-tuning. Ultimately, MinerU2.5 demonstrates strong document parsing ability, achieving state-of-the-art performance on multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lower computational overhead.

Linqing Wang, Ximing Xing, Yiji Cheng et al.

Recent advancements in text-to-image (T2I) diffusion models have demonstrated remarkable capabilities in generating high-fidelity images. However, these models often struggle to faithfully render complex user prompts, particularly in aspects like attribute binding, negation, and compositional relationships. This leads to a significant mismatch between user intent and the generated output. To address this challenge, we introduce PromptEnhancer, a novel and universal prompt rewriting framework that enhances any pretrained T2I model without requiring modifications to its weights. Unlike prior methods that rely on model-specific fine-tuning or implicit reward signals like image-reward scores, our framework decouples the rewriter from the generator. We achieve this by training a Chain-of-Thought (CoT) rewriter through reinforcement learning, guided by a dedicated reward model we term the AlignEvaluator. The AlignEvaluator is trained to provide explicit and fine-grained feedback based on a systematic taxonomy of 24 key points, which are derived from a comprehensive analysis of common T2I failure modes. By optimizing the CoT rewriter to maximize the reward from our AlignEvaluator, our framework learns to generate prompts that are more precisely interpreted by T2I models. Extensive experiments on the HunyuanImage 2.1 model demonstrate that PromptEnhancer significantly improves image-text alignment across a wide range of semantic and compositional challenges. Furthermore, we introduce a new, high-quality human preference benchmark to facilitate future research in this direction.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Open-vocabulary segmentation aims to identify and segment specific regions and objects based on text-based descriptions. A common solution is to leverage powerful vision-language models (VLMs), such as CLIP, to bridge the gap between vision and text information. However, VLMs are typically pretrained for image-level vision-text alignment, focusing on global semantic features. In contrast, segmentation tasks require fine-grained pixel-level alignment and detailed category boundary information, which VLMs alone cannot provide. As a result, information extracted directly from VLMs can't meet the requirements of segmentation tasks. To address this limitation, we propose FGAseg, a model designed for fine-grained pixel-text alignment and category boundary supplementation. The core of FGAseg is a Pixel-Level Alignment module that employs a cross-modal attention mechanism and a text-pixel alignment loss to refine the coarse-grained alignment from CLIP, achieving finer-grained pixel-text semantic alignment. Additionally, to enrich category boundary information, we introduce the alignment matrices as optimizable pseudo-masks during forward propagation and propose Category Information Supplementation module. These pseudo-masks, derived from cosine and convolutional similarity, provide essential global and local boundary information between different categories. By combining these two strategies, FGAseg effectively enhances pixel-level alignment and category boundary information, addressing key challenges in open-vocabulary segmentation. Extensive experiments demonstrate that FGAseg outperforms existing methods on open-vocabulary semantic segmentation benchmarks.

Bingyu Li, Feiyu Wang, Da Zhang et al.

Most existing underwater instance segmentation approaches are constrained by close-vocabulary prediction, limiting their ability to recognize novel marine categories. To support evaluation, we introduce \textbf{MARIS} (\underline{Mar}ine Open-Vocabulary \underline{I}nstance \underline{S}egmentation), the first large-scale fine-grained benchmark for underwater Open-Vocabulary (OV) segmentation, featuring a limited set of seen categories and diverse unseen categories. Although OV segmentation has shown promise on natural images, our analysis reveals that transfer to underwater scenes suffers from severe visual degradation (e.g., color attenuation) and semantic misalignment caused by lack underwater class definitions. To address these issues, we propose a unified framework with two complementary components. The Geometric Prior Enhancement Module (\textbf{GPEM}) leverages stable part-level and structural cues to maintain object consistency under degraded visual conditions. The Semantic Alignment Injection Mechanism (\textbf{SAIM}) enriches language embeddings with domain-specific priors, mitigating semantic ambiguity and improving recognition of unseen categories. Experiments show that our framework consistently outperforms existing OV baselines both In-Domain and Cross-Domain setting on MARIS, establishing a strong foundation for future underwater perception research.

Haoyang Li, Huan Gao, Zhiyuan Zhao et al.

The widespread adoption of Large Language Models (LLMs) has heightened concerns about their security, particularly their vulnerability to jailbreak attacks that leverage crafted prompts to generate malicious outputs. While prior research has been conducted on general security capabilities of LLMs, their specific susceptibility to jailbreak attacks in code generation remains largely unexplored. To fill this gap, we propose MalwareBench, a benchmark dataset containing 3,520 jailbreaking prompts for malicious code-generation, designed to evaluate LLM robustness against such threats. MalwareBench is based on 320 manually crafted malicious code generation requirements, covering 11 jailbreak methods and 29 code functionality categories. Experiments show that mainstream LLMs exhibit limited ability to reject malicious code-generation requirements, and the combination of multiple jailbreak methods further reduces the model's security capabilities: specifically, the average rejection rate for malicious content is 60.93%, dropping to 39.92% when combined with jailbreak attack algorithms. Our work highlights that the code security capabilities of LLMs still pose significant challenges.

Da Zhang, Bingyu Li, Zhiyuan Zhao et al.

Time series classification (TSC) is crucial in numerous real-world applications, such as environmental monitoring, medical diagnosis, and posture recognition. TSC tasks require models to effectively capture discriminative information for accurate class identification. Although deep learning architectures excel at capturing temporal dependencies, they often suffer from high computational cost, sensitivity to noise perturbations, and susceptibility to overfitting on small-scale datasets. To address these challenges, we propose FAIM, a lightweight Frequency-Aware Interactive Mamba model. Specifically, we introduce an Adaptive Filtering Block (AFB) that leverages Fourier Transform to extract frequency-domain features from time series data. The AFB incorporates learnable adaptive thresholds to dynamically suppress noise and employs element-wise coupling of global and local semantic adaptive filtering, enabling in-depth modeling of the synergy among different frequency components. Furthermore, we design an Interactive Mamba Block (IMB) to facilitate efficient multi-granularity information interaction, balancing the extraction of fine-grained discriminative features and comprehensive global contextual information, thereby endowing FAIM with powerful and expressive representations for TSC tasks. Additionally, we incorporate a self-supervised pre-training mechanism to enhance FAIM's understanding of complex temporal patterns and improve its robustness across various domains and high-noise scenarios. Extensive experiments on multiple benchmarks demonstrate that FAIM consistently outperforms existing state-of-the-art (SOTA) methods, achieving a superior trade-off between accuracy and efficiency and exhibits outstanding performance.

Zhiyuan Zhao, Haoxin Liu, B. Aditya Prakash

Time-series forecasting finds broad applications in real-world scenarios. Due to the dynamic nature of time series data, it is important for time-series forecasting models to handle potential distribution shifts over time. In this paper, we initially identify two types of distribution shifts in time series: concept drift and temporal shift. We acknowledge that while existing studies primarily focus on addressing temporal shift issues in time series forecasting, designing proper concept drift methods for time series forecasting has received comparatively less attention. Motivated by the need to address potential concept drift, while conventional concept drift methods via invariant learning face certain challenges in time-series forecasting, we propose a soft attention mechanism that finds invariant patterns from both lookback and horizon time series. Additionally, we emphasize the critical importance of mitigating temporal shifts as a preliminary to addressing concept drift. In this context, we introduce ShifTS, a method-agnostic framework designed to tackle temporal shift first and then concept drift within a unified approach. Extensive experiments demonstrate the efficacy of ShifTS in consistently enhancing the forecasting accuracy of agnostic models across multiple datasets, and outperforming existing concept drift, temporal shift, and combined baselines.

Zhiyuan Zhao, Yubin Wen, Siyu Yang et al.

Crowd counting is a task of estimating the number of the crowd through images, which is extremely valuable in the fields of intelligent security, urban planning, public safety management, and so on. However, the existing counting methods have some problems in practical application on embedded systems for these fields, such as excessive model parameters, abundant complex calculations, etc. The practical application of embedded systems requires the model to be real-time, which means that the model is fast enough. Considering the aforementioned problems, we design a super real-time model with a stem-encoder-decoder structure for crowd counting tasks, which achieves the fastest inference compared with state-of-the-arts. Firstly, large convolution kernels in the stem network are used to enlarge the receptive field, which effectively extracts detailed head information. Then, in the encoder part, we use conditional channel weighting and multi-branch local fusion block to merge multi-scale features with low computational consumption. This part is crucial to the super real-time performance of the model. Finally, the feature pyramid networks are added to the top of the encoder to alleviate its incomplete fusion problems. Experiments on three benchmarks show that our network is suitable for super real-time crowd counting on embedded systems, ensuring competitive accuracy. At the same time, the proposed network reasoning speed is the fastest. Specifically, the proposed network achieves 381.7 FPS on NVIDIA GTX 1080Ti and 71.9 FPS on NVIDIA Jetson TX1.