PDF 理解

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.

Current large multimodal models (LMMs) face significant challenges in processing and comprehending long-duration or high-resolution videos, which is mainly due to the lack of high-quality datasets. To address this issue from a data-centric perspective, we propose VISTA, a simple yet effective video spatiotemporal augmentation framework that synthesizes long-duration and high-resolution video instruction-following pairs from existing video-caption datasets. VISTA spatially and temporally combines videos to create new synthetic videos with extended durations and enhanced resolutions, and subsequently produces question-answer pairs pertaining to these newly synthesized videos. Based on this paradigm, we develop seven video augmentation methods and curate VISTA-400K, a video instruction-following dataset aimed at enhancing long-duration and high-resolution video understanding. Finetuning various video LMMs on our data resulted in an average improvement of 3.3% across four challenging benchmarks for long-video understanding. Furthermore, we introduce the first comprehensive high-resolution video understanding benchmark HRVideoBench, on which our finetuned models achieve a 6.5% performance gain. These results highlight the effectiveness of our framework.

With the rapid development of Multi-modal Large Language Models (MLLMs), an increasing number of benchmarks have been established to evaluate the video understanding capabilities of these models. However, these benchmarks focus on standalone videos and only assess "visual elements" like human actions and object states. In reality, contemporary videos often encompass complex and continuous narratives, typically presented as a series. To address this challenge, we propose SeriesBench, a benchmark consisting of 105 carefully curated narrative-driven series, covering 28 specialized tasks that require deep narrative understanding to solve. Specifically, we first select a diverse set of drama series spanning various genres. Then, we introduce a novel long-span narrative annotation method, combined with a full-information transformation approach to convert manual annotations into diverse task formats. To further enhance the model's capacity for detailed analysis of plot structures and character relationships within series, we propose a novel narrative reasoning framework, PC-DCoT. Extensive results on SeriesBench indicate that existing MLLMs still face significant challenges in understanding narrative-driven series, while PC-DCoT enables these MLLMs to achieve performance improvements. Overall, our SeriesBench and PC-DCoT highlight the critical necessity of advancing model capabilities for understanding narrative-driven series, guiding future MLLMs development. SeriesBench is publicly available at https://github.com/zackhxn/SeriesBench-CVPR2025.

Vision-Language Models (VLMs) are crucial for real-world applications that require understanding textual and visual information. However, existing VLMs face multiple challenges in processing long videos, including computational inefficiency, memory limitations, and difficulties maintaining coherent understanding across extended sequences. These issues stem partly from the quadratic scaling of self-attention w.r.t. number of tokens but also encompass broader challenges in temporal reasoning and information integration over long sequences. To address these challenges, we introduce ReWind, a novel two-stage framework for long video understanding. In the first stage, ReWind maintains a dynamic memory that stores and updates instruction-relevant visual information as the video unfolds.Memory updates leverage novel read and write mechanisms utilizing learnable queries and cross-attentions between memory contents and the input stream. This approach maintains low memory requirements as the cross-attention layers scale linearly w.r.t. number of tokens. In the second stage, the memory content guides the selection of a few relevant frames, represented at high spatial resolution, which are combined with the memory contents and fed into an LLM to generate the final answer. We empirically demonstrate ReWind's superiority in visual question answering (VQA) and temporal grounding tasks, surpassing previous methods on long video benchmarks. Notably, ReWind achieves a +13% score gain and a +12% accuracy improvement on the MovieChat-1K VQA dataset and an +8% mIoU increase on Charades-STA for temporal grounding.

Efficient understanding of long-form videos remains a significant challenge in computer vision. In this work, we revisit temporal search paradigms for long-form video understanding, studying a fundamental issue pertaining to all state-of-the-art (SOTA) long-context vision-language models (VLMs). In particular, our contributions are two-fold: **First**, we formulate temporal search as a **Long Video Haystack** problem, i.e., finding a minimal set of relevant frames (typically one to five) among tens of thousands of frames from real-world long videos given specific queries. To validate our formulation, we create **LV-Haystack**, the first benchmark containing 3,874 human-annotated instances with fine-grained evaluation metrics for assessing keyframe search quality and computational efficiency. Experimental results on LV-Haystack highlight a significant research gap in temporal search capabilities, with SOTA keyframe selection methods achieving only 2.1% temporal F1 score on the LVBench subset.**Next**, inspired by visual search in images, we re-think temporal searching and propose a lightweight keyframe searching framework, T^* , which casts the expensive temporal search as a spatial search problem. T^* leverages superior visual localization capabilities typically used in images and introduces an adaptive zooming-in mechanism that operates across both temporal and spatial dimensions. Our extensive experiments show that when integrated with existing methods, T^* significantly improves SOTA long-form video understanding performance. Specifically, under an inference budget of 32 frames, T^* improves GPT-4o's performance from 50.5% to **53.1%** and LLaVA-OneVision-72B's performance from 56.5% to **62.4%** on LongVideoBench XL subset. Our PyTorch code, benchmark dataset and models are included in the Supplementary material.

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.

Multi-modal Large Language Models (MLLMs) have introduced a novel dimension to document understanding, i.e., they endow large language models with visual comprehension capabilities; however, how to design a suitable image-text pre-training task for bridging the visual and language modality in document-level MLLMs remains underexplored. In this study, we introduce a novel visual-language alignment method that casts the key issue as a Visual Question Answering with Mask generation (VQAMask) task, optimizing two tasks simultaneously: VQA-based text parsing and mask generation. The former allows the model to implicitly align images and text at the semantic level. The latter introduces an additional mask generator (discarded during inference) to explicitly ensure alignment between visual texts within images and their corresponding image regions at a spatially-aware level. Together, they can prevent model hallucinations when parsing visual text and effectively promote spatially-aware feature representation learning. To support the proposed VQAMask task, we construct a comprehensive image-mask generation pipeline and provide a large-scale dataset with 6M data (MTMask6M). Subsequently, we demonstrate that introducing the proposed mask generation task yields competitive document-level understanding performance. Leveraging the proposed VQAMask, we introduce Marten, a training-efficient MLLM tailored for document-level understanding. Extensive experiments show that our Marten consistently achieves significant improvements among 8B-MLLMs in document-centric tasks. Code and datasets are available at https://github.com/PriNing/Marten.

How can we enable models to comprehend video anomalies occurring over varying temporal scales and contexts?Traditional Video Anomaly Understanding (VAU) methods focus on frame-level anomaly prediction, often missing the interpretability of complex and diverse real-world anomalies. Recent multimodal approaches leverage visual and textual data but lack hierarchical annotations that capture both short-term and long-term anomalies.To address this challenge, we introduce HIVAU-70k, a large-scale benchmark for hierarchical video anomaly understanding across any granularity. We develop a semi-automated annotation engine that efficiently scales high-quality annotations by combining manual video segmentation with recursive free-text annotation using large language models (LLMs). This results in over 70,000 multi-granular annotations organized at clip-level, event-level, and video-level segments.For efficient anomaly detection in long videos, we propose the Anomaly-focused Temporal Sampler (ATS). ATS integrates an anomaly scorer with a density-aware sampler to adaptively select frames based on anomaly scores, ensuring that the multimodal LLM concentrates on anomaly-rich regions, which significantly enhances both efficiency and accuracy.Extensive experiments demonstrate that our hierarchical instruction data markedly improves anomaly comprehension. The integrated ATS and visual-language model outperform traditional methods in processing long videos.Our benchmark and model will be publicly available.

We introduce Fish-Visual Trait Analysis (Fish-Vista), the first organismal image dataset designed for the analysis of visual traits of aquatic species directly from images using machine learning and computer vision methods. Fish-Vista contains 69,269 annotated images spanning 4,316 fish species, curated and organized to serve three downstream tasks: species classification, trait identification, and trait segmentation. Our work makes two key contributions. First, we provide a fully reproducible data processing pipeline to process fish images sourced from various museum collections, contributing to the advancement of AI in biodiversity science. We annotate the images with carefully curated labels from biological databases and manual annotations to create an AI-ready dataset of visual traits. Second, our work offers fertile grounds for researchers to develop novel methods for a variety of problems in computer vision such as handling long-tailed distributions, out-of-distribution generalization, learning with weak labels, explainable AI, and segmenting small objects. Dataset and code for Fish-Vista are available at https://github.com/Imageomics/Fish-Vista

The challenge in LLM-based video understanding lies in preserving visual and semantic information in long videos while maintaining a memory-affordable token count. However, redundancy and correspondence in videos have hindered the performance potential of existing methods. Through statistical learning on current datasets, we observe that redundancy occurs in both repeated and answer-irrelevant frames, and the corresponding frames vary with different questions. This suggests the possibility of adopting dynamic encoding to balance detailed video information preservation with token budget reduction. To this end, we propose a dynamic cooperative network, DynFocus, for memory-efficient video encoding in this paper. Specifically, i) a Dynamic Event Prototype Estimation (DPE) module to dynamically select meaningful frames for question answering; (ii) a Compact Cooperative Encoding (CCE) module that encodes meaningful frames with detailed visual appearance and the remaining frames with sketchy perception separately. We evaluate our method on five publicly available benchmarks, and experimental results consistently demonstrate that our method achieves competitive performance.

Most of the existing methods for video understanding primarily focus on videos only lasting tens of seconds, with limited exploration of techniques for handling long videos. The increased number of frames in long videos poses two main challenges: difficulty in locating key information and performing long-range reasoning. Thus, we propose DrVideo, a document-retrieval-based system designed for long video understanding. Our key idea is to convert the long-video understanding problem into a long-document understanding task so as to effectively leverage the power of large language models. Specifically, DrVideo first transforms a long video into a coarse text-based long document to initially retrieve key frames and then updates the documents with the augmented key frame information. It then employs an agent-based iterative loop to continuously search for missing information and augment the document until sufficient question-related information is gathered for making the final predictions in a chain-of-thought manner. Extensive experiments on long video benchmarks confirm the effectiveness of our method. DrVideo significantly outperforms existing LLM-based state-of-the-art methods on EgoSchema benchmark (3 minutes), MovieChat-1K benchmark (10 minutes), and the long split of Video-MME benchmark (average of 44 minutes). Code is available at https://github.com/Upper9527/DrVideo.

Despite significant progress in multimodal large language models (MLLMs), their performance on complex, multi-page document comprehension remains inadequate, largely due to the lack of high-quality, document-level datasets. While current retrieval-augmented generation (RAG) methods offer partial solutions, they suffer from issues, such as fragmented retrieval contexts, multi-stage error accumulation, and extra time costs of retrieval. In this work, we present a high-quality document-level dataset, Doc-750K, designed to support in-depth understanding of multimodal documents. This dataset includes diverse document structures, extensive cross-page dependencies, and real question-answer pairs derived from the original documents. Building on the dataset, we develop a native multimodal model--Docopilot, which can accurately handle document-level dependencies without relying on RAG. Experiments demonstrate that Docopilot achieves superior coherence, accuracy, and efficiency in document understanding tasks and multi-turn interactions, setting a new baseline for document-level multimodal understanding. Data, code, and models are released at https://github.com/OpenGVLab/Docopilot.

Text-rich document understanding (TDU) requires comprehensive analysis of documents containing substantial textual content and complex layouts. While Multimodal Large Language Models (MLLMs) have achieved fast progress in this domain, existing approaches either demand significant computational resources or struggle with effective multi-modal integration. In this paper, we introduce DocLayLLM, an efficient multi-modal extension of LLMs specifically designed for TDU. By lightly integrating visual patch tokens and 2D positional tokens into LLMs' input and encoding the document content using the LLMs themselves, we fully take advantage of the document comprehension capability of LLMs and enhance their perception of OCR information. We have also deeply considered the role of chain-of-thought (CoT) and innovatively proposed the techniques of CoT Pre-training and CoT Annealing. Our DocLayLLM can achieve remarkable performances with lightweight training settings, showcasing its efficiency and effectiveness. Experimental results demonstrate that our DocLayLLM outperforms existing OCR-dependent methods and OCR-free competitors. Code and model are available at https://github.com/whlscut/DocLayLLM.

Large video-language models (VLMs) have demonstrated promising progress in various video understanding tasks. However, their effectiveness in long-form video analysis is constrained by limited context windows. Traditional approaches, such as uniform frame sampling, often inevitably allocate resources to irrelevant content, diminishing their effectiveness in real-world scenarios. In this paper, we introduce BOLT, a method to boost large VLMs without additional training through a comprehensive study of frame selection strategies. First, to enable a more realistic evaluation of VLMs in long-form video understanding, we propose a multi-source retrieval evaluation setting. Our findings reveal that uniform sampling performs poorly in noisy contexts, underscoring the importance of selecting the right frames. Second, we explore several frame selection strategies based on query-frame similarity and analyze their effectiveness at inference time. Our results show that inverse transform sampling yields the most significant performance improvement, increasing accuracy on the Video-MME benchmark from 53.8% to 56.1% and MLVU benchmark from 58.9% to 63.4%.

Visual Document Understanding has become essential with the increase of text-rich visual content. This field poses significant challenges due to the need for effective integration of visual perception and textual comprehension, particularly across diverse document types with complex layouts. Moreover, existing fine-tuning datasets for this domain often fall short in providing the detailed contextual information for robust understanding, leading to hallucinations and limited comprehension of spatial relationships among visual elements. To address these challenges, we propose an innovative pipeline that utilizes adaptive generation of markup languages, such as Markdown, JSON, HTML, and TiKZ, to build highly structured document representations and deliver contextually-grounded responses. We introduce two fine-grained structured datasets: DocMark-Pile, comprising approximately 3.8M pretraining data pairs for document parsing, and DocMark-Instruct, featuring 624k fine-tuning data annotations for grounded instruction following.Extensive experiments demonstrate that our proposed model significantly outperforms existing state-of-the-art MLLMs across a range of visual document understanding benchmarks, facilitating advanced reasoning and comprehension capabilities in complex visual scenarios.

Multimodal large language models (MLLMs) have enabled open-world visual understanding by injecting visual input as extra tokens into large language models (LLMs) as contexts. However, when the visual input changes from a single image to a long video, the above paradigm encounters difficulty because the vast amount of video tokens has significantly exceeded the maximal capacity of MLLMs. Therefore, existing video-based MLLMs are mostly established upon sampling a small portion of tokens from input data, which can cause key information to be lost and thus produce incorrect answers. This paper presents a simple yet effective algorithm named Adaptive Keyframe Sampling (AKS). It inserts a plug-and-play module known as keyframe selection, which aims to maximize the useful information with a fixed number of video tokens. We formulate keyframe selection as an optimization involving (1) the relevance between the keyframes and the prompt, and (2) the coverage of the keyframes over the video, and present an adaptive algorithm to approximate the best solution. Experiments on two long video understanding benchmarks validate that AKS improves video QA accuracy (beyond strong baselines) upon selecting informative keyframes. Our study reveals the importance of information pre-filtering in video-based MLLMs. Our code are available at https://github.com/ncTimTang/AKS

The advancements in large language models (LLMs) have propelled the improvement of video understanding tasks by incorporating LLMs with visual models. However, most existing LLM-based models (e.g., VideoLLaMA, VideoChat) are constrained to processing short-duration videos. Recent attempts to understand long-term videos by extracting and compressing visual features into a fixed memory size. Nevertheless, those methods leverage only visual modality to merge video tokens and overlook the correlation between visual and textual queries, leading to difficulties in effectively handling complex question-answering tasks. To address the challenges of long videos and complex prompts, we propose AdaCM^2, which, for the first time, introduces an adaptive cross-modality memory reduction approach to video-text alignment in an auto-regressive manner on video streams. Our extensive experiments on various video understanding tasks, such as video captioning, video question answering, and video classification, demonstrate that AdaCM^2 achieves state-of-the-art performance across multiple datasets while significantly reducing memory usage. Notably, it achieves a 4.5% improvement across multiple tasks in the LVU dataset with a GPU memory consumption reduction of up to 65%.

Recent methods that integrate spatial layouts with text for document understanding in large language models (LLMs) have shown promising results. A commonly used method is to represent layout information as text tokens and interleave them with text content as inputs to the LLMs. However, such a method still demonstrates limitations, as it requires additional position IDs for tokens that are used to represent layout information. Due to the constraint on max position IDs, assigning them to layout information reduces those available for text content, reducing the capacity for the model to learn from the text during training, while also introducing a large number of potentially untrained position IDs during long-context inference, which can hinder performance on document understanding tasks. To address these issues, we propose LayTokenLLM, a simple yet effective method for document understanding. LayTokenLLM represents layout information as a single token per text segment and uses a specialized positional encoding scheme. It shares position IDs between text and layout tokens, eliminating the need for additional position IDs. This design maintains the model's capacity to learn from text while mitigating long-context issues during inference. Furthermore, a novel pre-training objective called Next Interleaved Text and Layout Token Prediction (NTLP) is devised to enhance cross-modality learning between text and layout tokens. Extensive experiments show that LayTokenLLM outperforms existing layout-integrated LLMs and MLLMs of similar scales on multi-page document understanding tasks, as well as most single-page tasks.