Sibo Song

Zhibo Yang, Rujiao Long, Pengfei Wang et al.

Recently, Visual Information Extraction (VIE) has been becoming increasingly important in both the academia and industry, due to the wide range of real-world applications. Previously, numerous works have been proposed to tackle this problem. However, the benchmarks used to assess these methods are relatively plain, i.e., scenarios with real-world complexity are not fully represented in these benchmarks. As the first contribution of this work, we curate and release a new dataset for VIE, in which the document images are much more challenging in that they are taken from real applications, and difficulties such as blur, partial occlusion, and printing shift are quite common. All these factors may lead to failures in information extraction. Therefore, as the second contribution, we explore an alternative approach to precisely and robustly extract key information from document images under such tough conditions. Specifically, in contrast to previous methods, which usually either incorporate visual information into a multi-modal architecture or train text spotting and information extraction in an end-to-end fashion, we explicitly model entities as semantic points, i.e., center points of entities are enriched with semantic information describing the attributes and relationships of different entities, which could largely benefit entity labeling and linking. Extensive experiments on standard benchmarks in this field as well as the proposed dataset demonstrate that the proposed method can achieve significantly enhanced performance on entity labeling and linking, compared with previous state-of-the-art models. Dataset is available at https://www.modelscope.cn/datasets/damo/SIBR/summary.

Sibo Song, Jianqiang Wan, Zhibo Yang et al.

Recently, vision-language joint representation learning has proven to be highly effective in various scenarios. In this paper, we specifically adapt vision-language joint learning for scene text detection, a task that intrinsically involves cross-modal interaction between the two modalities: vision and language, since text is the written form of language. Concretely, we propose to learn contextualized, joint representations through vision-language pre-training, for the sake of enhancing the performance of scene text detectors. Towards this end, we devise a pre-training architecture with an image encoder, a text encoder and a cross-modal encoder, as well as three pretext tasks: image-text contrastive learning (ITC), masked language modeling (MLM) and word-in-image prediction (WIP). The pre-trained model is able to produce more informative representations with richer semantics, which could readily benefit existing scene text detectors (such as EAST and PSENet) in the down-stream text detection task. Extensive experiments on standard benchmarks demonstrate that the proposed paradigm can significantly improve the performance of various representative text detectors, outperforming previous pre-training approaches. The code and pre-trained models will be publicly released.

Mingxin Li, Yanzhao Zhang, Dingkun Long et al.

In this report, we introduce the Qwen3-VL-Embedding and Qwen3-VL-Reranker model series, the latest extensions of the Qwen family built on the Qwen3-VL foundation model. Together, they provide an end-to-end pipeline for high-precision multimodal search by mapping diverse modalities, including text, images, document images, and video, into a unified representation space. The Qwen3-VL-Embedding model employs a multi-stage training paradigm, progressing from large-scale contrastive pre-training to reranking model distillation, to generate semantically rich high-dimensional vectors. It supports Matryoshka Representation Learning, enabling flexible embedding dimensions, and handles inputs up to 32k tokens. Complementing this, Qwen3-VL-Reranker performs fine-grained relevance estimation for query-document pairs using a cross-encoder architecture with cross-attention mechanisms. Both model series inherit the multilingual capabilities of Qwen3-VL, supporting more than 30 languages, and are released in $\textbf{2B}$ and $\textbf{8B}$ parameter sizes to accommodate diverse deployment requirements. Empirical evaluations demonstrate that the Qwen3-VL-Embedding series achieves state-of-the-art results across diverse multimodal embedding evaluation benchmarks. Specifically, Qwen3-VL-Embedding-8B attains an overall score of $\textbf{77.8}$ on MMEB-V2, ranking first among all models (as of January 8, 2025). This report presents the architecture, training methodology, and practical capabilities of the series, demonstrating their effectiveness on various multimodal retrieval tasks, including image-text retrieval, visual question answering, and video-text matching.

Jianqiang Wan, Sibo Song, Wenwen Yu et al.

Recently, visually-situated text parsing (VsTP) has experienced notable advancements, driven by the increasing demand for automated document understanding and the emergence of Generative Large Language Models (LLMs) capable of processing document-based questions. Various methods have been proposed to address the challenging problem of VsTP. However, due to the diversified targets and heterogeneous schemas, previous works usually design task-specific architectures and objectives for individual tasks, which inadvertently leads to modal isolation and complex workflow. In this paper, we propose a unified paradigm for parsing visually-situated text across diverse scenarios. Specifically, we devise a universal model, called OmniParser, which can simultaneously handle three typical visually-situated text parsing tasks: text spotting, key information extraction, and table recognition. In OmniParser, all tasks share the unified encoder-decoder architecture, the unified objective: point-conditioned text generation, and the unified input & output representation: prompt & structured sequences. Extensive experiments demonstrate that the proposed OmniParser achieves state-of-the-art (SOTA) or highly competitive performances on 7 datasets for the three visually-situated text parsing tasks, despite its unified, concise design. The code is available at https://github.com/AlibabaResearch/AdvancedLiterateMachinery.

Wenwen Yu, Zhibo Yang, Jianqiang Wan et al.

Visually-situated text parsing (VsTP) has recently seen notable advancements, driven by the growing demand for automated document understanding and the emergence of large language models capable of processing document-based questions. While various methods have been proposed to tackle the complexities of VsTP, existing solutions often rely on task-specific architectures and objectives for individual tasks. This leads to modal isolation and complex workflows due to the diversified targets and heterogeneous schemas. In this paper, we introduce OmniParser V2, a universal model that unifies VsTP typical tasks, including text spotting, key information extraction, table recognition, and layout analysis, into a unified framework. Central to our approach is the proposed Structured-Points-of-Thought (SPOT) prompting schemas, which improves model performance across diverse scenarios by leveraging a unified encoder-decoder architecture, objective, and input\&output representation. SPOT eliminates the need for task-specific architectures and loss functions, significantly simplifying the processing pipeline. Our extensive evaluations across four tasks on eight different datasets show that OmniParser V2 achieves state-of-the-art or competitive results in VsTP. Additionally, we explore the integration of SPOT within a multimodal large language model structure, further enhancing text localization and recognition capabilities, thereby confirming the generality of SPOT prompting technique. The code is available at \href{https://github.com/AlibabaResearch/AdvancedLiterateMachinery}{AdvancedLiterateMachinery}.

Songtao Jiang, Sibo Song, Chenyi Zhou et al.

The transition from image to video understanding requires vision-language models (VLMs) to shift from recognizing static patterns to reasoning over temporal dynamics such as motion trajectories, speed changes, and state transitions. Yet current post-training methods fall short due to two critical limitations: (1) existing datasets often lack temporal-centricity, where answers can be inferred from isolated keyframes rather than requiring holistic temporal integration; and (2) training data generated by proprietary models contains systematic errors in fundamental temporal perception, such as confusing motion directions or misjudging speeds. We introduce SynRL, a post-training framework that teaches models temporal primitives, the fundamental building blocks of temporal understanding including direction, speed, and state tracking. Our key insight is that these abstract primitives, learned from programmatically generated synthetic videos, transfer effectively to real-world scenarios. We decompose temporal understanding into short-term perceptual primitives (speed, direction) and long-term cognitive primitives, constructing 7.7K CoT and 7K RL samples with ground-truth frame-level annotations through code-based video generation. Despite training on simple geometric shapes, SynRL achieves substantial improvements across 15 benchmarks spanning temporal grounding, complex reasoning, and general video understanding. Remarkably, our 7.7K synthetic CoT samples outperform Video-R1 with 165K real-world samples. We attribute this to fundamental temporal skills, such as tracking frame by frame changes and comparing velocity, that transfer effectively from abstract synthetic patterns to complex real-world scenarios. This establishes a new paradigm for video post-training: video temporal learning through carefully designed synthetic data provides a more cost efficient scaling path.

Shuai Bai, Keqin Chen, Xuejing Liu et al. · pku

We introduce Qwen2.5-VL, the latest flagship model of Qwen vision-language series, which demonstrates significant advancements in both foundational capabilities and innovative functionalities. Qwen2.5-VL achieves a major leap forward in understanding and interacting with the world through enhanced visual recognition, precise object localization, robust document parsing, and long-video comprehension. A standout feature of Qwen2.5-VL is its ability to localize objects using bounding boxes or points accurately. It provides robust structured data extraction from invoices, forms, and tables, as well as detailed analysis of charts, diagrams, and layouts. To handle complex inputs, Qwen2.5-VL introduces dynamic resolution processing and absolute time encoding, enabling it to process images of varying sizes and videos of extended durations (up to hours) with second-level event localization. This allows the model to natively perceive spatial scales and temporal dynamics without relying on traditional normalization techniques. By training a native dynamic-resolution Vision Transformer (ViT) from scratch and incorporating Window Attention, we reduce computational overhead while maintaining native resolution. As a result, Qwen2.5-VL excels not only in static image and document understanding but also as an interactive visual agent capable of reasoning, tool usage, and task execution in real-world scenarios such as operating computers and mobile devices. Qwen2.5-VL is available in three sizes, addressing diverse use cases from edge AI to high-performance computing. The flagship Qwen2.5-VL-72B model matches state-of-the-art models like GPT-4o and Claude 3.5 Sonnet, particularly excelling in document and diagram understanding. Additionally, Qwen2.5-VL maintains robust linguistic performance, preserving the core language competencies of the Qwen2.5 LLM.

Songtao Jiang, Yuan Wang, Sibo Song et al.

Real-world clinical decision-making requires integrating heterogeneous data, including medical text, 2D images, 3D volumes, and videos, while existing AI systems fail to unify all these signals, limiting their utility. In this paper, we introduce Hulu-Med, a transparent, generalist medical Vision-Language Model (VLM) designed to unify language-only, 2D/3D vision-language, and video understanding within a single architecture. Hulu-Med is trained on a curated corpus of 16.7 million samples, comprising exclusively public or synthetic data, spanning 12 major anatomical systems and 14 medical imaging modalities. Hulu-Med employs a medical-aware token-reduction strategy that prunes redundant visual tokens, achieving up to a 55% reduction for 3D and video inputs, improving cross-modal efficiency, and enabling training at 7B-32B parameter scales in approximately 4,000-40,000 GPU hours. Across 30 public in-domain and out-of-domain medical benchmarks-covering text reasoning, visual question answering, report generation, multilingual dialogue, video understanding, and rare disease diagnosis-Hulu-Med surpasses existing open-source models on 27 of 30 benchmarks and outperforms proprietary systems such as GPT-4o on 16 benchmarks. Despite being a VLM, Hulu-Med outperforms GPT-4o and matches GPT-o1 on the text-only HealthBench. For the first time in the community, we provide a fully transparent, reproducible and cost-effective pipeline for holistic medical vision-language understanding by releasing our end-to-end data curation, training procedures, and model parameters. Code and models are available at https://github.com/ZJUI-AI4H/Hulu-Med.

Jie Huang, Xuejing Liu, Sibo Song et al.

Multimodal position encoding is essential for vision-language models, yet there has been little systematic investigation into multimodal position encoding. We conduct a comprehensive analysis of multimodal Rotary Positional Embedding (RoPE) by examining its two core components: position design and frequency allocation. Through extensive experiments, we identify three key guidelines: positional coherence, full frequency utilization, and preservation of textual priors-ensuring unambiguous layout, rich representation, and faithful transfer from the pre-trained LLM. Based on these insights, we propose Multi-Head RoPE (MHRoPE) and MRoPE-Interleave (MRoPE-I), two simple and plug-and-play variants that require no architectural changes. Our methods consistently outperform existing approaches across diverse benchmarks, with significant improvements in both general and fine-grained multimodal understanding. Code will be avaliable at https://github.com/JJJYmmm/Multimodal-RoPEs.

Songtao Jiang, Yuan Wang, Sibo Song et al.

The practical deployment of medical vision-language models (Med-VLMs) necessitates seamless integration of textual data with diverse visual modalities, including 2D/3D images and videos, yet existing models typically employ separate encoders for different modalities. To address this limitation, we present OmniV-Med, a unified framework for multimodal medical understanding. Our technical contributions are threefold: First, we construct OmniV-Med-Instruct, a comprehensive multimodal medical dataset containing 252K instructional samples spanning 14 medical image modalities and 11 clinical tasks. Second, we devise a rotary position-adaptive encoder that processes multi-resolution 2D/3D images and videos within a unified architecture, diverging from conventional modality-specific encoders. Third, we introduce a medical-aware token pruning mechanism that exploits spatial-temporal redundancy in volumetric data (e.g., consecutive CT slices) and medical videos, effectively reducing 60\% of visual tokens without performance degradation. Empirical evaluations demonstrate that OmniV-Med-7B achieves state-of-the-art performance on 7 benchmarks spanning 2D/3D medical imaging and video understanding tasks. Notably, our lightweight variant (OmniV-Med-1.5B) attains comparable performance while requiring only 8 RTX3090 GPUs for training and supporting efficient long-video inference. Data, code and model will be released.

Songtao Jiang, Yuan Wang, Ruizhe Chen et al.

In medical visual question answering (Med-VQA), achieving accurate responses relies on three critical steps: precise perception of medical imaging data, logical reasoning grounded in visual input and textual questions, and coherent answer derivation from the reasoning process. Recent advances in general vision-language models (VLMs) show that large-scale reinforcement learning (RL) could significantly enhance both reasoning capabilities and overall model performance. However, their application in medical domains is hindered by two fundamental challenges: 1) misalignment between perceptual understanding and reasoning stages, and 2) inconsistency between reasoning pathways and answer generation, both compounded by the scarcity of high-quality medical datasets for effective large-scale RL. In this paper, we first introduce Med-Zero-17K, a curated dataset for pure RL-based training, encompassing over 30 medical image modalities and 24 clinical tasks. Moreover, we propose a novel large-scale RL framework for Med-VLMs, Consistency-Aware Preference Optimization (CAPO), which integrates rewards to ensure fidelity between perception and reasoning, consistency in reasoning-to-answer derivation, and rule-based accuracy for final responses. Extensive experiments on both in-domain and out-of-domain scenarios demonstrate the superiority of our method over strong VLM baselines, showcasing strong generalization capability to 3D Med-VQA benchmarks and R1-like training paradigms.

Shuai Bai, Yuxuan Cai, Ruizhe Chen et al.

We introduce Qwen3-VL, the most capable vision-language model in the Qwen series to date, achieving superior performance across a broad range of multimodal benchmarks. It natively supports interleaved contexts of up to 256K tokens, seamlessly integrating text, images, and video. The model family includes both dense (2B/4B/8B/32B) and mixture-of-experts (30B-A3B/235B-A22B) variants to accommodate diverse latency-quality trade-offs. Qwen3-VL delivers three core pillars: (i) markedly stronger pure-text understanding, surpassing comparable text-only backbones in several cases; (ii) robust long-context comprehension with a native 256K-token window for both text and interleaved multimodal inputs, enabling faithful retention, retrieval, and cross-referencing across long documents and videos; and (iii) advanced multimodal reasoning across single-image, multi-image, and video tasks, demonstrating leading performance on comprehensive evaluations such as MMMU and visual-math benchmarks (e.g., MathVista and MathVision). Architecturally, we introduce three key upgrades: (i) an enhanced interleaved-MRoPE for stronger spatial-temporal modeling across images and video; (ii) DeepStack integration, which effectively leverages multi-level ViT features to tighten vision-language alignment; and (iii) text-based time alignment for video, evolving from T-RoPE to explicit textual timestamp alignment for more precise temporal grounding. Under comparable token budgets and latency constraints, Qwen3-VL achieves superior performance in both dense and Mixture-of-Experts (MoE) architectures. We envision Qwen3-VL serving as a foundational engine for image-grounded reasoning, agentic decision-making, and multimodal code intelligence in real-world workflows.

Songtao Jiang, Yuxi Chen, Sibo Song et al.

In high-stakes medical applications, consistent answering across diverse question phrasings is essential for reliable diagnosis. However, we reveal that current Medical Vision-Language Models (Med-VLMs) exhibit concerning fragility in Medical Visual Question Answering, as their answers fluctuate significantly when faced with semantically equivalent rephrasings of medical questions. We attribute this to two limitations: (1) insufficient alignment of medical concepts, leading to divergent reasoning patterns, and (2) hidden biases in training data that prioritize syntactic shortcuts over semantic understanding. To address these challenges, we construct RoMed, a dataset built upon original VQA datasets containing 144k questions with variations spanning word-level, sentence-level, and semantic-level perturbations. When evaluating state-of-the-art (SOTA) models like LLaVA-Med on RoMed, we observe alarming performance drops (e.g., a 40\% decline in Recall) compared to original VQA benchmarks, exposing critical robustness gaps. To bridge this gap, we propose Consistency and Contrastive Learning (CCL), which integrates two key components: (1) knowledge-anchored consistency learning, aligning Med-VLMs with medical knowledge rather than shallow feature patterns, and (2) bias-aware contrastive learning, mitigating data-specific priors through discriminative representation refinement. CCL achieves SOTA performance on three popular VQA benchmarks and notably improves answer consistency by 50\% on the challenging RoMed test set, demonstrating significantly enhanced robustness. Code will be released.

Zhibo Yang, Wei Hua, Sibo Song et al.

Visual Information Extraction (VIE), aiming at extracting structured information from visually rich document images, plays a pivotal role in document processing. Considering various layouts, semantic scopes, and languages, VIE encompasses an extensive range of types, potentially numbering in the thousands. However, many of these types suffer from a lack of training data, which poses significant challenges. In this paper, we propose a novel generative model, named Generative Compositor, to address the challenge of few-shot VIE. The Generative Compositor is a hybrid pointer-generator network that emulates the operations of a compositor by retrieving words from the source text and assembling them based on the provided prompts. Furthermore, three pre-training strategies are employed to enhance the model's perception of spatial context information. Besides, a prompt-aware resampler is specially designed to enable efficient matching by leveraging the entity-semantic prior contained in prompts. The introduction of the prompt-based retrieval mechanism and the pre-training strategies enable the model to acquire more effective spatial and semantic clues with limited training samples. Experiments demonstrate that the proposed method achieves highly competitive results in the full-sample training, while notably outperforms the baseline in the 1-shot, 5-shot, and 10-shot settings.

Sibo Song, Ngai-Man Cheung, Vijay Chandrasekhar et al.

Deep neural networks have recently achieved competitive accuracy for human activity recognition. However, there is room for improvement, especially in modeling long-term temporal importance and determining the activity relevance of different temporal segments in a video. To address this problem, we propose a learnable and differentiable module: Deep Adaptive Temporal Pooling (DATP). DATP applies a self-attention mechanism to adaptively pool the classification scores of different video segments. Specifically, using frame-level features, DATP regresses importance of different temporal segments and generates weights for them. Remarkably, DATP is trained using only the video-level label. There is no need of additional supervision except video-level activity class label. We conduct extensive experiments to investigate various input features and different weight models. Experimental results show that DATP can learn to assign large weights to key video segments. More importantly, DATP can improve training of frame-level feature extractor. This is because relevant temporal segments are assigned large weights during back-propagation. Overall, we achieve state-of-the-art performance on UCF101, HMDB51 and Kinetics datasets.

Sibo Song, Yueru Chen, Ngai-Man Cheung et al.

Deep neural networks (DNNs) are known to be vulnerable to adversarial perturbations, which imposes a serious threat to DNN-based decision systems. In this paper, we propose to apply the lossy Saak transform to adversarially perturbed images as a preprocessing tool to defend against adversarial attacks. Saak transform is a recently-proposed state-of-the-art for computing the spatial-spectral representations of input images. Empirically, we observe that outputs of the Saak transform are very discriminative in differentiating adversarial examples from clean ones. Therefore, we propose a Saak transform based preprocessing method with three steps: 1) transforming an input image to a joint spatial-spectral representation via the forward Saak transform, 2) apply filtering to its high-frequency components, and, 3) reconstructing the image via the inverse Saak transform. The processed image is found to be robust against adversarial perturbations. We conduct extensive experiments to investigate various settings of the Saak transform and filtering functions. Without harming the decision performance on clean images, our method outperforms state-of-the-art adversarial defense methods by a substantial margin on both the CIFAR-10 and ImageNet datasets. Importantly, our results suggest that adversarial perturbations can be effectively and efficiently defended using state-of-the-art frequency analysis.

Zhe Wang, Kingsley Kuan, Mathieu Ravaut et al.

The YouTube-8M video classification challenge requires teams to classify 0.7 million videos into one or more of 4,716 classes. In this Kaggle competition, we placed in the top 3% out of 650 participants using released video and audio features. Beyond that, we extend the original competition by including text information in the classification, making this a truly multi-modal approach with vision, audio and text. The newly introduced text data is termed as YouTube-8M-Text. We present a classification framework for the joint use of text, visual and audio features, and conduct an extensive set of experiments to quantify the benefit that this additional mode brings. The inclusion of text yields state-of-the-art results, e.g. 86.7% GAP on the YouTube-8M-Text validation dataset.

Yiren Zhou, Sibo Song, Ngai-Man Cheung

Image blur and image noise are common distortions during image acquisition. In this paper, we systematically study the effect of image distortions on the deep neural network (DNN) image classifiers. First, we examine the DNN classifier performance under four types of distortions. Second, we propose two approaches to alleviate the effect of image distortion: re-training and fine-tuning with noisy images. Our results suggest that, under certain conditions, fine-tuning with noisy images can alleviate much effect due to distorted inputs, and is more practical than re-training.

Sibo Song, Ngai-Man Cheung, Vijay Chandrasekhar et al.

With the increasing availability of wearable devices, research on egocentric activity recognition has received much attention recently. In this paper, we build a Multimodal Egocentric Activity dataset which includes egocentric videos and sensor data of 20 fine-grained and diverse activity categories. We present a novel strategy to extract temporal trajectory-like features from sensor data. We propose to apply the Fisher Kernel framework to fuse video and temporal enhanced sensor features. Experiment results show that with careful design of feature extraction and fusion algorithm, sensor data can enhance information-rich video data. We make publicly available the Multimodal Egocentric Activity dataset to facilitate future research.