Yan Shu

Niantong Li, Guangzheng Hu, Weixu Qiao et al.

Text-to-Image generation has evolved from basic image synthesis into a frequently used core capability in professional creative workflows, where simple text-image alignment can no longer satisfy users' pressing demands for faithful real-world reconstruction and genuine creative expression. Existing benchmarks, however, remain anchored in these foundational criteria and do not yet capture the nuanced capabilities that matter in authentic artistic practice, making it difficult to reliably distinguish state-of-the-art T2I models. To address the gap, we introduce Qwen-Image-Bench, a creator-centric benchmark co-designed with professional artists and grounded in real-world creation scenarios. Qwen-Image-Bench enriches conventional evaluation with two application-driven dimensions: Real-world Fidelity and Creative Generation. Drawing on the staged reasoning inherent in professional artistic workflows, we organize these five pillars into a top-down hierarchical taxonomy that further decomposes into 23 second-level sub-capabilities and 56 third-level verifiable rubrics. To ensure broad coverage, we curate 1000 stratified prompts with each prompt jointly exercising more than four fine-grained facets across multiple pillars. We train a unified judge model Q-Judger based on Qwen3.6-27B, supervised by 80 professional annotators from global art academies under blind labeling and triple-review protocols, that scores every image across all 56 verifiable facets, producing fine-grained, rubric-grounded, and fully attributable diagnostics rather than a single opaque score. Empirically, Qwen-Image-Bench reliably distinguishes leading T2I models, achieving the greatest separation on the two application-driven dimensions of Real-world Fidelity and Creative Generation where existing benchmarks provide little insight, while also providing a trustworthy optimization signal for production-level T2I development.

Tianhe Wu, Kun Yan, Zikai Zhou et al.

Few-step distillation has become an effective strategy for accelerating advanced visual generative models, yet prior work has largely focused on distillation objectives. In this work, we revisit few-step distillation from a complementary perspective, focusing on the training recipe that critically shapes student performance. Using Qwen-Image-2.0 as a representative case, we systematically investigate three factors in unified text-to-image generation and instruction-guided image editing distillation: data composition, teacher guidance, and task mixture. Our empirical analysis reveals several non-obvious behaviors, which motivate the development of Qwen-Image-Flash. Overall, our results suggest that effective few-step distillation requires not only carefully designed objectives, but also principled organization of the broader training pipeline.

Yan Shu, Zheng Liu, Peitian Zhang et al.

Long video understanding poses a significant challenge for current Multi-modal Large Language Models (MLLMs). Notably, the MLLMs are constrained by their limited context lengths and the substantial costs while processing long videos. Although several existing methods attempt to reduce visual tokens, their strategies encounter severe bottleneck, restricting MLLMs' ability to perceive fine-grained visual details. In this work, we propose Video-XL, a novel approach that leverages MLLMs' inherent key-value (KV) sparsification capacity to condense the visual input. Specifically, we introduce a new special token, the Visual Summarization Token (VST), for each interval of the video, which summarizes the visual information within the interval as its associated KV. The VST module is trained by instruction fine-tuning, where two optimizing strategies are offered. 1.Curriculum learning, where VST learns to make small (easy) and large compression (hard) progressively. 2. Composite data curation, which integrates single-image, multi-image, and synthetic data to overcome the scarcity of long-video instruction data. The compression quality is further improved by dynamic compression, which customizes compression granularity based on the information density of different video intervals. Video-XL's effectiveness is verified from three aspects. First, it achieves a superior long-video understanding capability, outperforming state-of-the-art models of comparable sizes across multiple popular benchmarks. Second, it effectively preserves video information, with minimal compression loss even at 16x compression ratio. Third, it realizes outstanding cost-effectiveness, enabling high-quality processing of thousands of frames on a single A100 GPU.

Zhengyang Liang, Yan Shu, Xiangrui Liu et al.

The evolution of autonomous agents is redefining information seeking, transitioning from passive retrieval to proactive, open-ended web research. However, a significant modality gap remains in processing the web's most dynamic and information-dense modality: video. In this paper, we first formalize the task of Agentic Video Browsing and introduce Video-BrowseComp, a benchmark evaluating open-ended agentic browsing tasks that enforce a mandatory dependency on videos. We observe that current paradigms struggle to reconcile the scale of open-ended video exploration with the need for fine-grained visual verification. Direct visual inference (e.g., RAG) maximizes perception but incurs prohibitive context costs, while text-centric summarization optimizes efficiency but often misses critical visual details required for accurate grounding. To address this, we propose Video-Browser, a novel agent leveraging Pyramidal Perception, filtering with cheap metadata and zooming in with expensive visual perception only when necessary. Experiments demonstrate that our approach achieves a 37.5% relative improvement while reducing token consumption by 58.3% compared to Direct visual inference, establishing a foundation for verifiable open-web video research. We open-source all codes, benchmark at {https://anonymous.4open.science/r/VideoBrowser} and {https://github.com/chrisx599/Video-Browser}.

Yachun Mi, Yu Li, Yan Shu et al.

Video Quality Assessment (VQA) aims to simulate the process of perceiving video quality by the human visual system (HVS). The judgments made by HVS are always influenced by human subjective feelings. However, most of the current VQA research focuses on capturing various distortions in the spatial and temporal domains of videos, while ignoring the impact of human feelings. In this paper, we propose CLiF-VQA, which considers both features related to human feelings and spatial features of videos. In order to effectively extract features related to human feelings from videos, we explore the consistency between CLIP and human feelings in video perception for the first time. Specifically, we design multiple objective and subjective descriptions closely related to human feelings as prompts. Further we propose a novel CLIP-based semantic feature extractor (SFE) which extracts features related to human feelings by sliding over multiple regions of the video frame. In addition, we further capture the low-level-aware features of the video through a spatial feature extraction module. The two different features are then aggregated thereby obtaining the quality score of the video. Extensive experiments show that the proposed CLiF-VQA exhibits excellent performance on several VQA datasets.

Yan Shu, Shaohui Liu, Honglei Xu et al.

Recently, developing an automatic reading system for analog measuring instruments has gained increased attention, as it enables the collection of numerous state of equipment. Nonetheless, two major obstacles still obstruct its deployment to real-world applications. The first issue is that they rarely take the entire pipeline's speed into account. The second is that they are incapable of dealing with some low-quality images (i.e., meter breakage, blur, and uneven scale). In this paper, we propose a human-like alignment and recognition algorithm to overcome these problems. More specifically, a Spatial Transformed Module(STM) is proposed to obtain the front view of images in a self-autonomous way based on an improved Spatial Transformer Networks(STN). Meanwhile, a Value Acquisition Module(VAM) is proposed to infer accurate meter values by an end-to-end trained framework. In contrast to previous research, our model aligns and recognizes meters totally implemented by learnable processing, which mimics human's behaviours and thus achieves higher performances. Extensive results verify the good robustness of the proposed model in terms of the accuracy and efficiency.

Zhenhang Li, Yan Shu, Weichao Zeng et al.

Diffusion models, known for their impressive image generation abilities, have played a pivotal role in the rise of visual text generation. Nevertheless, existing visual text generation methods often focus on generating entire images with text prompts, leading to imprecise control and limited practicality. A more promising direction is visual text blending, which focuses on seamlessly merging texts onto text-free backgrounds. However, existing visual text blending methods often struggle to generate high-fidelity and diverse images due to a shortage of backgrounds for synthesis and limited generalization capabilities. To overcome these challenges, we propose a new visual text blending paradigm including both creating backgrounds and rendering texts. Specifically, a background generator is developed to produce high-fidelity and text-free natural images. Moreover, a text renderer named GlyphOnly is designed for achieving visually plausible text-background integration. GlyphOnly, built on a Stable Diffusion framework, utilizes glyphs and backgrounds as conditions for accurate rendering and consistency control, as well as equipped with an adaptive text block exploration strategy for small-scale text rendering. We also explore several downstream applications based on our method, including scene text dataset synthesis for boosting scene text detectors, as well as text image customization and editing. Code and model will be available at \url{https://github.com/Zhenhang-Li/GlyphOnly}.

Minghao Qin, Xiangrui Liu, Zhengyang Liang et al.

Multi-modal large language models (MLLMs) models have made significant progress in video understanding over the past few years. However, processing long video inputs remains a major challenge due to high memory and computational costs. This makes it difficult for current models to achieve both strong performance and high efficiency in long video understanding. To address this challenge, we propose Video-XL-2, a novel MLLM that delivers superior cost-effectiveness for long-video understanding based on task-aware KV sparsification. The proposed framework operates with two key steps: chunk-based pre-filling and bi-level key-value decoding. Chunk-based pre-filling divides the visual token sequence into chunks, applying full attention within each chunk and sparse attention across chunks. This significantly reduces computational and memory overhead. During decoding, bi-level key-value decoding selectively reloads either dense or sparse key-values for each chunk based on its relevance to the task. This approach further improves memory efficiency and enhances the model's ability to capture fine-grained information. Video-XL-2 achieves state-of-the-art performance on various long video understanding benchmarks, outperforming existing open-source lightweight models. It also demonstrates exceptional efficiency, capable of processing over 10,000 frames on a single NVIDIA A100 (80GB) GPU and thousands of frames in just a few seconds.

Yan Shu, Weichao Zeng, Fangmin Zhao et al.

Visual text is a crucial component in both document and scene images, conveying rich semantic information and attracting significant attention in the computer vision community. Beyond traditional tasks such as text detection and recognition, visual text processing has witnessed rapid advancements driven by the emergence of foundation models, including text image reconstruction and text image manipulation. Despite significant progress, challenges remain due to the unique properties that differentiate text from general objects. Effectively capturing and leveraging these distinct textual characteristics is essential for developing robust visual text processing models. In this survey, we present a comprehensive, multi-perspective analysis of recent advancements in visual text processing, focusing on two key questions: (1) What textual features are most suitable for different visual text processing tasks? (2) How can these distinctive text features be effectively incorporated into processing frameworks? Furthermore, we introduce VTPBench, a new benchmark that encompasses a broad range of visual text processing datasets. Leveraging the advanced visual quality assessment capabilities of multimodal large language models (MLLMs), we propose VTPScore, a novel evaluation metric designed to ensure fair and reliable evaluation. Our empirical study with more than 20 specific models reveals substantial room for improvement in the current techniques. Our aim is to establish this work as a fundamental resource that fosters future exploration and innovation in the dynamic field of visual text processing. The relevant repository is available at https://github.com/shuyansy/Visual-Text-Processing-survey.

Yan Shu, Bin Ren, Zhitong Xiong et al.

Vision-language models (VLMs) have shown promise in earth observation (EO), yet they struggle with tasks that require grounding complex spatial reasoning in precise pixel-level visual representations. To address this problem, we introduce TerraScope, a unified VLM that delivers pixel-grounded geospatial reasoning with two key capabilities: (1) modality-flexible reasoning: it handles single-modality inputs (optical or SAR) and adaptively fuses different modalities into the reasoning process when both are available; (2) multi-temporal reasoning: it integrates temporal sequences for change analysis across multiple time points. In addition, we curate Terra-CoT, a large-scale dataset containing 1 million samples with pixel-level masks embedded in reasoning chains across multiple sources. We also propose TerraScope-Bench, the first benchmark for pixel-grounded geospatial reasoning with six sub-tasks that evaluates both answer accuracy and mask quality to ensure authentic pixel-grounded reasoning. Experiments show that TerraScope significantly outperforms existing VLMs on pixel-grounded geospatial reasoning while providing interpretable visual evidence.

Zeyu Chen, Fangmin Zhao, Yan Shu et al.

Style-conditioned scene text generation faces unique challenges in extracting precise text styles from complex backgrounds and maintaining fine-grained style consistency across characters, especially for multilingual scripts. We propose StyleTextGen, a novel framework that learns to perceive and replicate visual text styles across different languages and writing systems. Our approach features three key contributions: First, we introduce a dual-branch style encoder dedicated to style modeling, yielding robust multilingual text style representations in complex real-world scenes. Second, we design a text style consistency loss that enhances style coherence and improves overall visual quality. Third, we develop a mask-guided inference strategy that ensures precise style alignment between generated and reference text. To facilitate systematic evaluation, we construct StyleText-CE, a bilingual scene text style benchmark covering both monolingual and cross-lingual settings. Extensive experiments demonstrate that StyleTextGen significantly outperforms existing methods in style consistency and cross-lingual generalization, establishing new state-of-the-art performance in multilingual style-conditioned text generation.

Bing Zhao, Chenfei Wu, Deqing Li et al.

We present Qwen-Image-2.0, an omni-capable image generation foundation model that unifies high-fidelity generation and precise image editing within a single framework. Despite recent progress, existing models still struggle with ultra-long text rendering, multilingual typography, high-resolution photorealism, robust instruction following, and efficient deployment, especially in text-rich and compositionally complex scenarios. Qwen-Image-2.0 addresses these challenges by coupling Qwen3-VL as the condition encoder with a Multimodal Diffusion Transformer for joint condition-target modeling, supported by large-scale data curation and a customized multi-stage training pipeline. This enables strong multimodal understanding while preserving flexible generation and editing capabilities. The model supports instructions of up to 1K tokens for generating text-rich content such as slides, posters, infographics, and comics, while significantly improving multilingual text fidelity and typography. It also enhances photorealistic generation with richer details, more realistic textures, and coherent lighting, and follows complex prompts more reliably across diverse styles. Extensive human evaluations show that Qwen-Image-2.0 substantially outperforms previous Qwen-Image models in both generation and editing, marking a step toward more general, reliable, and practical image generation foundation models.

Zekai Zhang, Deqing Li, Kuan Cao et al.

We present Qwen-Image-VAE-2.0, a suite of high-compression Variational Autoencoders (VAEs) that achieve significant advances in both reconstruction fidelity and diffusability. To address the reconstruction bottlenecks of high compression, we adopt an improved architecture featuring Global Skip Connections (GSC) and expanded latent channels. Moreover, we scale training to billions of images and incorporate a synthetic rendering engine to improve performance in text-rich scenarios. To tackle the convergence challenges of high-dimensional latent space, we implement an enhanced semantic alignment strategy to make the latent space highly amenable to diffusion modeling. To optimize computational efficiency, we leverage an asymmetric and attention-free encoder-decoder backbone to minimize encoding overhead. We present a comprehensive evaluation of Qwen-Image-VAE-2.0 on public reconstruction benchmarks. To evaluate performance in text-rich scenarios, we propose OmniDoc-TokenBench, a new benchmark comprising a diverse collection of real-world documents coupled with specialized OCR-based evaluation metrics. Qwen-Image-VAE-2.0 achieves state-of-the-art reconstruction performance, demonstrating exceptional capabilities in both general domains and text-rich scenarios at high compression ratio. Furthermore, downstream DiT experiments reveal our models possess superior diffusability, significantly accelerating convergence compared to existing high-compression baselines. These establish Qwen-Image-VAE-2.0 as a leading model with high compression, superior reconstruction, and exceptional diffusability.

Chi Liu, Derek Li, Yan Shu et al.

While large language models show promise in medical applications, achieving expert-level clinical reasoning remains challenging due to the need for both accurate answers and transparent reasoning processes. To address this challenge, we introduce Fleming-R1, a model designed for verifiable medical reasoning through three complementary innovations. First, our Reasoning-Oriented Data Strategy (RODS) combines curated medical QA datasets with knowledge-graph-guided synthesis to improve coverage of underrepresented diseases, drugs, and multi-hop reasoning chains. Second, we employ Chain-of-Thought (CoT) cold start to distill high-quality reasoning trajectories from teacher models, establishing robust inference priors. Third, we implement a two-stage Reinforcement Learning from Verifiable Rewards (RLVR) framework using Group Relative Policy Optimization, which consolidates core reasoning skills while targeting persistent failure modes through adaptive hard-sample mining. Across diverse medical benchmarks, Fleming-R1 delivers substantial parameter-efficient improvements: the 7B variant surpasses much larger baselines, while the 32B model achieves near-parity with GPT-4o and consistently outperforms strong open-source alternatives. These results demonstrate that structured data design, reasoning-oriented initialization, and verifiable reinforcement learning can advance clinical reasoning beyond simple accuracy optimization. We release Fleming-R1 publicly to promote transparent, reproducible, and auditable progress in medical AI, enabling safer deployment in high-stakes clinical environments.

Huaying Yuan, Zheng Liu, Junjie Zhou et al.

Long-video understanding~(LVU) is a challenging problem in computer vision. Existing methods either downsample frames for single-pass reasoning, sacrificing fine-grained details, or depend on textual reasoning over task-agnostic representations, hindering task-specific perception and exploration. In this paper, we propose VideoExplorer, a framework grounded in the principle of ``thinking with video'', which naturally intertwines planning, temporal grounding, and scalable perception into a coherent reasoning process. Rather than reasoning over a static context, VideoExplorer iteratively formulates sub-questions, locates relevant moments, and performs task-oriented, temporally scalable video understanding until reaching the final answer, enabling faithful, efficient, and interpretable reasoning. To address the lack of LVU training resources, we construct a long-video reasoning dataset using difficulty-adaptive sampling to ensure high-quality trajectories on complex tasks. Building on this dataset, we design a two-stage training pipeline: supervised trajectory initialization followed by trajectory-level preference optimization, encouraging adaptive temporal grounding and iterative information integration guided by downstream rewards. Extensive evaluations on popular long-video understanding and reasoning benchmarks demonstrate VideoExplorer's significant advantage over existing baselines, highlighting its robustness, adaptability, and efficiency. Our code is made publicly available in this repository(https://github.com/yhy-2000/VideoDeepResearch).

Bin Ren, Hang Guo, Yan Shu et al.

This paper reviews the NTIRE 2026 challenge on efficient single-image super-resolution with a focus on the proposed solutions and results. The aim of this challenge is to devise a network that reduces one or several aspects, such as runtime, parameters, and FLOPs, while maintaining PSNR of around 26.90 dB on the DIV2K_LSDIR_valid dataset, and 26.99 dB on the DIV2K_LSDIR_test dataset. The challenge had 95 registered participants, and 15 teams made valid submissions. They gauge the state-of-the-art results for efficient single-image super-resolution.

Xiangrui Liu, Yan Shu, Zheng Liu et al.

Despite advanced token compression techniques, existing multimodal large language models (MLLMs) still struggle with hour-long video understanding. In this work, we propose Video-XL-Pro, an efficient method for extremely long video understanding, built upon Reconstructive Compression of Tokens (ReCoT), a learnable module that leverages self-supervised learning to generate comprehensive and compact video tokens. ReCoT introduces two key components: (i) Dynamic Token Synthesizer (DTS): DTS generates pseudo-video tokens from static image tokens by learning intra-token relationships, which are then used in masked video modeling. (ii) Semantic-Guided Masking (SGM): SGM adaptively masks redundant visual tokens to facilitate more effective reconstructive learning. To improve training efficiency in MLLMs fine-tuning, we introduce a video-specific dataset pruning strategy and design a simple yet Query-aware Selector that enables the model to precisely locate query-relevant video tokens. With only 3B parameters, Video-XL-Pro outperforms most 7B models trained on larger datasets across multiple long video understanding benchmarks. Moreover, it can process over 8K frames on a single A100 GPU while maintaining high-quality performance.

Weichao Zeng, Yan Shu, Zhenhang Li et al.

Centred on content modification and style preservation, Scene Text Editing (STE) remains a challenging task despite considerable progress in text-to-image synthesis and text-driven image manipulation recently. GAN-based STE methods generally encounter a common issue of model generalization, while Diffusion-based STE methods suffer from undesired style deviations. To address these problems, we propose TextCtrl, a diffusion-based method that edits text with prior guidance control. Our method consists of two key components: (i) By constructing fine-grained text style disentanglement and robust text glyph structure representation, TextCtrl explicitly incorporates Style-Structure guidance into model design and network training, significantly improving text style consistency and rendering accuracy. (ii) To further leverage the style prior, a Glyph-adaptive Mutual Self-attention mechanism is proposed which deconstructs the implicit fine-grained features of the source image to enhance style consistency and vision quality during inference. Furthermore, to fill the vacancy of the real-world STE evaluation benchmark, we create the first real-world image-pair dataset termed ScenePair for fair comparisons. Experiments demonstrate the effectiveness of TextCtrl compared with previous methods concerning both style fidelity and text accuracy.

Huaying Yuan, Zheng Liu, Minghao Qin et al.

Efficient long-video understanding~(LVU) remains a challenging task in computer vision. Current long-context vision-language models~(LVLMs) suffer from information loss due to compression and brute-force downsampling. While retrieval-augmented generation (RAG) methods mitigate this issue, their applicability is limited due to explicit query dependency. To overcome this challenge, we introduce a novel memory-enhanced RAG-based approach called MemVid, which is inspired by the cognitive memory of human beings. Our approach operates in four basic steps: 1) memorizing holistic video information, 2) reasoning about the task's information needs based on memory, 3) retrieving critical moments based on the information needs, and 4) focusing on the retrieved moments to produce the final answer. To enhance the system's memory-grounded reasoning capabilities while achieving optimal end-to-end performance, we propose a curriculum learning strategy. This approach begins with supervised learning on well-annotated reasoning results, then progressively explores and reinforces more plausible reasoning outcomes through reinforcement learning. We perform extensive evaluations on popular LVU benchmarks, including MLVU, VideoMME and LVBench. In our experiments, MemVid demonstrates superior efficiency and effectiveness compared to both LVLMs and RAG methods.

Yan Shu, Weichao Zeng, Zhenhang Li et al.

Visual text, a pivotal element in both document and scene images, speaks volumes and attracts significant attention in the computer vision domain. Beyond visual text detection and recognition, the field of visual text processing has experienced a surge in research, driven by the advent of fundamental generative models. However, challenges persist due to the unique properties and features that distinguish text from general objects. Effectively leveraging these unique textual characteristics is crucial in visual text processing, as observed in our study. In this survey, we present a comprehensive, multi-perspective analysis of recent advancements in this field. Initially, we introduce a hierarchical taxonomy encompassing areas ranging from text image enhancement and restoration to text image manipulation, followed by different learning paradigms. Subsequently, we conduct an in-depth discussion of how specific textual features such as structure, stroke, semantics, style, and spatial context are seamlessly integrated into various tasks. Furthermore, we explore available public datasets and benchmark the reviewed methods on several widely-used datasets. Finally, we identify principal challenges and potential avenues for future research. Our aim is to establish this survey as a fundamental resource, fostering continued exploration and innovation in the dynamic area of visual text processing.

Yan Shu, Chi Liu, Robin Chen et al.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable effectiveness in various general-domain scenarios, such as visual question answering and image captioning. Recently, researchers have increasingly focused on empowering MLLMs with medical conversational abilities, which hold significant promise for clinical applications. However, medical data presents unique challenges due to its heterogeneous nature -- encompassing diverse modalities including 2D images, 3D volumetric scans, and temporal video sequences. The substantial domain gap and data format inconsistencies across these modalities have hindered the development of unified medical MLLMs. To address these challenges, we propose Fleming-VL, a unified end-to-end framework for comprehensive medical visual understanding across heterogeneous modalities. Fleming-VL tackles this problem from a data-centric perspective through three key strategies: (1) scaling up pretraining by integrating long-context data from both natural and medical-specific domains; (2) complementing fine-tuning with rare medical data, including holistic video analysis and underrepresented 2D modalities such as ultrasound and dermoscopy images; (3) extending existing evaluation frameworks to incorporate 3D volumetric and video understanding benchmarks. Through supervised fine-tuning (SFT) and group relative policy optimization (GRPO), we develop Fleming-VL in multiple model scales. Extensive experiments demonstrate that Fleming-VL achieves state-of-the-art performance across multiple benchmarks, including medical VQA, video QA, and 3D medical image understanding. We publicly release Fleming-VL to promote transparent, reproducible, and auditable progress in medical AI.

Zhoufaran Yang, Yan Shu, Jing Wang et al. · pku

Visual texts embedded in videos carry rich semantic information, which is crucial for both holistic video understanding and fine-grained reasoning about local human actions. However, existing video understanding benchmarks largely overlook textual information, while OCR-specific benchmarks are constrained to static images, limiting their ability to capture the interaction between text and dynamic visual contexts. To address this gap, we propose VidText, a new benchmark designed for comprehensive and in-depth evaluation of video text understanding. VidText offers the following key features: 1) It covers a wide range of real-world scenarios and supports multilingual content, encompassing diverse settings where video text naturally appears. 2) It introduces a hierarchical evaluation framework with video-level, clip-level, and instance-level tasks, enabling assessment of both global summarization and local retrieval capabilities. 3) The benchmark also introduces a set of paired perception reasoning tasks, ranging from visual text perception to cross-modal reasoning between textual and visual information. Extensive experiments on 18 state-of-the-art Large Multimodal Models (LMMs) reveal that current models struggle across most tasks, with significant room for improvement. Further analysis highlights the impact of both model-intrinsic factors, such as input resolution and OCR capability, and external factors, including the use of auxiliary information and Chain-of-Thought reasoning strategies. We hope VidText will fill the current gap in video understanding benchmarks and serve as a foundation for future research on multimodal reasoning with video text in dynamic environments.

Yan Shu, Hangui Lin, Yexin Liu et al.

Large Multimodal Models (LMMs) have achieved impressive progress in visual perception and reasoning. However, when confronted with visually ambiguous or non-semantic scene text, they often struggle to accurately spot and understand the content, frequently generating semantically plausible yet visually incorrect answers, which we refer to as semantic hallucination. In this work, we investigate the underlying causes of semantic hallucination and identify a key finding: Transformer layers in LLM with stronger attention focus on scene text regions are less prone to producing semantic hallucinations. Thus, we propose a training-free semantic hallucination mitigation framework comprising two key components: (1) ZoomText, a coarse-to-fine strategy that identifies potential text regions without external detectors; and (2) Grounded Layer Correction, which adaptively leverages the internal representations from layers less prone to hallucination to guide decoding, correcting hallucinated outputs for non-semantic samples while preserving the semantics of meaningful ones. To enable rigorous evaluation, we introduce TextHalu-Bench, a benchmark of 1,740 samples spanning both semantic and non-semantic cases, with manually curated question answer pairs designed to probe model hallucinations. Extensive experiments demonstrate that our method not only effectively mitigates semantic hallucination but also achieves strong performance on public benchmarks for scene text spotting and understanding.

Johann-Ludwig Herzog, Mathis Jürgen Adler, Leonard Hackel et al.

Vision-langugage models (VLMs) have shown strong performance in computer vision (CV), yet their performance on remote sensing (RS) data remains limited due to the lack of large-scale, multi-sensor RS image-text datasets with diverse textual annotations. Existing datasets predominantly include aerial Red-Green-Blue imagery, with short or weakly grounded captions, and provide limited diversity in annotation types. To address this limitation, we introduce BigEarthNet$.$txt, a large-scale, multi-sensor image-text dataset designed to advance instruction-driven image-text learning in Earth observation across multiple tasks. BigEarthNet$.$txt contains 464044 co-registered Sentinel-1 synthetic aperture radar and Sentinel-2 multispectral images with 9.6M text annotations, including: i) geographically anchored captions describing land-use/land-cover (LULC) classes, their spatial relations, and environmental context; ii) visual question answering pairs relevant for different tasks; and iii) referring expression detection instructions for bounding box prediction. Through a comparative statistical analysis, we demonstrate that BigEarthNet$.$txt surpasses existing RS image-text datasets in textual richness and annotation type variety. We further establish a manually-verified benchmark split to evaluate VLMs in RS and CV. The results show the limitations of these models on tasks that involve complex LULC classes, whereas fine-tuning using BigEarthNet$.$txt results in consistent performance gains across all considered tasks.

Minghao Qin, Yan Shu, Peitian Zhang et al.

Long-video understanding (LVU) remains a severe challenge for existing multimodal large language models (MLLMs), primarily due to the prohibitive computational cost. Recent approaches have explored KV compression to mitigate this issue, but they often suffer from significant information loss at high compression ratios. In this paper, we introduce Video-X^2L, which flexibly preserves critical video information for each LVU task. Video-X^2L involves two key operations. The first one is called bi-level KV compression. During the MLLM's pre-filling stage, Video-X^2L generates two types of compressed KVs: low-compression KVs (L-KVs) to capture fine-grained video details and high-compression KVs (H-KVs) to offer compact video representations. The second one is called selective KV re-loading. During the MLLM's decoding stage, Video-X^2L selectively re-loads L-KVs for the most critical video chunks while using H-KVs for other less important ones. This allows the MLLM to fully utilize task-specific information while maintaining the overall compactness. Video-X^2L is simple yet effective: it is free from additional training and directly compatible with existing KV-compressible MLLMs. We evaluate Video-X^2L with a variety of popular LVU benchmarks, including VideoMME, MLVU, LongVideoBench, and VNBench. Our experiment result shows that Video-X^2L outperforms existing KV-compression methods by a huge advantage while substantially saving the computation cost.

Honglei Xu, Yan Shu, Shaohui Liu

Single image dehazing is a challenging ill-posed problem. Existing datasets for training deep learning-based methods can be generated by hand-crafted or synthetic schemes. However, the former often suffers from small scales, while the latter forces models to learn scene depth instead of haze distribution, decreasing their dehazing ability. To overcome the problem, we propose a simple yet novel synthetic method to decouple the relationship between haze density and scene depth, by which a depth-agnostic dataset (DA-HAZE) is generated. Meanwhile, a Global Shuffle Strategy (GSS) is proposed for generating differently scaled datasets, thereby enhancing the generalization ability of the model. Extensive experiments indicate that models trained on DA-HAZE achieve significant improvements on real-world benchmarks, with less discrepancy between SOTS and DA-SOTS (the test set of DA-HAZE). Additionally, Depth-agnostic dehazing is a more complicated task because of the lack of depth prior. Therefore, an efficient architecture with stronger feature modeling ability and fewer computational costs is necessary. We revisit the U-Net-based architectures for dehazing, in which dedicatedly designed blocks are incorporated. However, the performances of blocks are constrained by limited feature fusion methods. To this end, we propose a Convolutional Skip Connection (CSC) module, allowing vanilla feature fusion methods to achieve promising results with minimal costs. Extensive experimental results demonstrate that current state-of-the-art methods. equipped with CSC can achieve better performance and reasonable computational expense, whether the haze distribution is relevant to the scene depth.

Xiangrui Liu, Minghao Qin, Yan Shu et al.

Identifying key moments in long videos is essential for downstream understanding and reasoning tasks. In this paper, we introduce a new problem, Taskoriented Temporal Grounding ToTG, which aims to localize time intervals containing the necessary information based on a task's natural description. Along with the definition, we also present ToTG Bench, a comprehensive benchmark for evaluating the performance on ToTG. ToTG is particularly challenging for traditional approaches due to their limited generalizability and difficulty in handling long videos. To address these challenges, we propose TimeScope, a novel framework built upon progressive reasoning. TimeScope first identifies a coarse-grained temporal scope in the long video that likely contains the key moments, and then refines this scope through finegrained moment partitioning. Additionally, we curate a highquality dataset, namely ToTG Pile, to enhance TimeScope's ability to perform progressive temporal grounding effectively. Extensive experiments demonstrate that TimeScope consistently outperforms both existing temporalgrounding methods and popular MLLMs across various settings, highlighting its effectiveness in addressing this new challenging problem.

Yan Shu, Bin Ren, Zhitong Xiong et al.

Earth Observation (EO) data analysis is vital for monitoring environmental and human dynamics. Recent Multimodal Large Language Models (MLLMs) show potential in EO understanding but remain restricted to single-sensor inputs, overlooking the complementarity across heterogeneous modalities. We propose EarthMind, a unified vision-language framework that handles both single- and cross-sensor inputs via an innovative hierarchical cross-modal attention (ie, HCA) design. Specifically, HCA hierarchically captures visual relationships across sensors and aligns them with language queries, enabling adaptive fusion of optical and Synthetic Aperture Radar (SAR) features. To support cross-sensor learning, we curate FusionEO, a 30K-pair dataset with diverse annotations, and establish EarthMind-Bench, a 2,841-pair benchmark with expert annotations for perception and reasoning tasks. Extensive experiments show that EarthMind achieves state-of-the-art results on EarthMind-Bench and surpasses existing MLLMs on multiple EO benchmarks.

Junjie Zhou, Yan Shu, Bo Zhao et al.

The evaluation of Long Video Understanding (LVU) performance poses an important but challenging research problem. Despite previous efforts, the existing video understanding benchmarks are severely constrained by several issues, especially the insufficient lengths of videos, a lack of diversity in video types and evaluation tasks, and the inappropriateness for evaluating LVU performances. To address the above problems, we propose a new benchmark called MLVU (Multi-task Long Video Understanding Benchmark) for the comprehensive and in-depth evaluation of LVU. MLVU presents the following critical values: \textit{1)} The substantial and flexible extension of video lengths, which enables the benchmark to evaluate LVU performance across a wide range of durations. \textit{2)} The inclusion of various video genres, e.g., movies, surveillance footage, egocentric videos, cartoons, game videos, etc., which reflects the models' LVU performances in different scenarios. \textit{3)} The development of diversified evaluation tasks, which enables a comprehensive examination of MLLMs' key abilities in long-video understanding. The empirical study with 23 latest MLLMs reveals significant room for improvement in today's technique, as all existing methods struggle with most of the evaluation tasks and exhibit severe performance degradation when handling longer videos. Additionally, it suggests that factors such as context length, image-understanding ability, and the choice of LLM backbone can play critical roles in future advancements. We anticipate that MLVU will advance the research of long video understanding by providing a comprehensive and in-depth analysis of MLLMs.

Zhaofan Qiu, Ting Yao, Yan Shu et al.

Video is complex due to large variations in motion and rich content in fine-grained visual details. Abstracting useful information from such information-intensive media requires exhaustive computing resources. This paper studies a two-step alternative that first condenses the video sequence to an informative "frame" and then exploits off-the-shelf image recognition system on the synthetic frame. A valid question is how to define "useful information" and then distill it from a video sequence down to one synthetic frame. This paper presents a novel Informative Frame Synthesis (IFS) architecture that incorporates three objective tasks, i.e., appearance reconstruction, video categorization, motion estimation, and two regularizers, i.e., adversarial learning, color consistency. Each task equips the synthetic frame with one ability, while each regularizer enhances its visual quality. With these, by jointly learning the frame synthesis in an end-to-end manner, the generated frame is expected to encapsulate the required spatio-temporal information useful for video analysis. Extensive experiments are conducted on the large-scale Kinetics dataset. When comparing to baseline methods that map video sequence to a single image, IFS shows superior performance. More remarkably, IFS consistently demonstrates evident improvements on image-based 2D networks and clip-based 3D networks, and achieves comparable performance with the state-of-the-art methods with less computational cost.

Martin Mihelich, Charles Dognin, Yan Shu et al.

Bagging can significantly improve the generalization performance of unstable machine learning algorithms such as trees or neural networks. Though bagging is now widely used in practice and many empirical studies have explored its behavior, we still know little about the theoretical properties of bagged predictions. In this paper, we theoretically investigate how the bagging method can reduce the Mean Squared Error (MSE) when applied on a statistical estimator. First, we prove that for any estimator, increasing the number of bagged estimators $N$ in the average can only reduce the MSE. This intuitive result, observed empirically and discussed in the literature, has not yet been rigorously proved. Second, we focus on the standard estimator of variance called unbiased sample variance and we develop an exact analytical expression of the MSE for this estimator with bagging. This allows us to rigorously discuss the number of iterations $N$ and the batch size $m$ of the bagging method. From this expression, we state that only if the kurtosis of the distribution is greater than $\frac{3}{2}$, the MSE of the variance estimator can be reduced with bagging. This result is important because it demonstrates that for distribution with low kurtosis, bagging can only deteriorate the performance of a statistical prediction. Finally, we propose a novel general-purpose algorithm to estimate with high precision the variance of a sample.

Davis Liang, Yan Shu

Multi-task learning (MTL) has recently contributed to learning better representations in service of various NLP tasks. MTL aims at improving the performance of a primary task, by jointly training on a secondary task. This paper introduces automated tasks, which exploit the sequential nature of the input data, as secondary tasks in an MTL model. We explore next word prediction, next character prediction, and missing word completion as potential automated tasks. Our results show that training on a primary task in parallel with a secondary automated task improves both the convergence speed and accuracy for the primary task. We suggest two methods for augmenting an existing network with automated tasks and establish better performance in topic prediction, sentiment analysis, and hashtag recommendation. Finally, we show that the MTL models can perform well on datasets that are small and colloquial by nature.