PDF 理解论文趋势 2023–2026

主要结论

• PDF 理解论文从 2023 年的 6 篇增长到 2026 年的 39 篇，三年间扩大到约 6.5 倍。
• CVPR 是发PDF 理解论文最多的会议，2023–2026 年共 39 篇，占四个会议总量的 38%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 17 篇。
• 在「文档与版式」的 5 个主题里，它的 2026 年论文数排第 1。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document Understanding
• ICML 2026Explainable Forensics of Manipulated Segments in Untrimmed Long Videos
• ICLR 2026Smooth Reading: Bridging the Gap of Recurrent LLM to Self-Attention LLM on Long-Context Understanding
• CVPR 2026MS-Temba Multi-Scale Temporal Mamba for Understanding Long Untrimmed Videos
• NeurIPS 2025Return of ChebNet: Understanding and Improving an Overlooked GNN on Long Range Tasks
• ICML 2025Position: Deep Learning is Not So Mysterious or Different
• ICLR 2025Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution
• CVPR 2025Fish-Vista: A Multi-Purpose Dataset for Understanding & Identification of Traits from Images
• NeurIPS 2024SRFUND: A Multi-Granularity Hierarchical Structure Reconstruction Benchmark in Form Understanding
• ICML 2024Towards Understanding the Word Sensitivity of Attention Layers: A Study via Random Features
• ICLR 2024Nougat: Neural Optical Understanding for Academic Documents
• CVPR 2024Action Scene Graphs for Long-Form Understanding of Egocentric Videos
• NeurIPS 2023Balance, Imbalance, and Rebalance: Understanding Robust Overfitting from a Minimax Game Perspective
• ICLR 2023Understanding Edge-of-Stability Training Dynamics with a Minimalist Example
• CVPR 2023Long Range Pooling for 3D Large-Scale Scene Understanding
• NeurIPS 2026Towards Understanding Momentum Acceleration in River-Valley Loss Landscape
• ICML 2026SCOUT: Active Information Foraging for Long-Text Understanding with Decoupled Epistemic States
• ICLR 2026Understanding the Learning Phases in Self-Supervised Learning via Critical Periods
• CVPR 2026Understanding Accelerating and Improving MeanFlow Training
• NeurIPS 2025PaceLLM: Brain-Inspired Large Language Models for Long-Context Understanding

全部PDF 理解论文

共 35 篇，第 1 / 1 页

1. NeurIPS 2026A Principled Optimal Transport Framework for Frame Selection in Long Video Understanding
2. NeurIPS 2026Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding
3. NeurIPS 2026Articulation in Prime: Primitive-Based Articulated Object Understanding from a Single Casual Video
4. NeurIPS 2026Beyond Flat Frames: Hierarchical Graph Reasoning for Long Video Understanding
5. NeurIPS 2026Diagnosing and Correcting Bias in MLLM for Long Video Understanding
6. NeurIPS 2026DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document Understanding
7. NeurIPS 2026Elephant in the Fridge: Constant-Memory Frame Packing for Long Video Understanding
8. NeurIPS 2026FrameRouter: Frame Budget Routing for Long-Video Understanding on Video-MME-v2 and Beyond
9. NeurIPS 2026FrameScout: Scouting Query-Relevant Frames for Long Video Understanding
10. NeurIPS 2026HIMMEL: Hierarchical Interleaved Multi-stream Motion Encoding for Long Video Understanding
11. NeurIPS 2026InfiniteVL: A Systematic Approach to Highly-Efficient, Ultra-Long Multimodal Understanding
12. NeurIPS 2026MedHorizon: Towards Long-context Medical Video Understanding in the Wild
13. NeurIPS 2026MEMEVO: A Memory-Evolved Video Agent for Long Video Understanding
14. NeurIPS 2026PDF-HR: Pose Distance Fields for Humanoid Robots
15. NeurIPS 2026PIVOT: A Unified Agentic Framework for Streaming Long-Video Understanding
16. NeurIPS 2026Q-Focus: Let the Question Guide What to See in Long Videos
17. NeurIPS 2026The VLM as Sensor: Bayesian Active Search for Long Video Understanding
18. NeurIPS 2026Time to Pay Attention! Understanding High Complexity Corpus Reasoning Tasks
19. NeurIPS 2026Towards Understanding Momentum Acceleration in River-Valley Loss Landscape
20. NeurIPS 2026Towards Understanding Self-Pretraining for Sequence Classification
21. NeurIPS 2026UTOPI: Efficient Egocentric Long-Video Understanding in AR via User-Guided Token Pre-Compression
22. NeurIPS 2025FlexSelect: Flexible Token Selection for Efficient Long Video Understanding
23. NeurIPS 2025NeedleInATable: Exploring Long-Context Capability of Large Language Models towards Long-Structured Tables
24. NeurIPS 2025PaceLLM: Brain-Inspired Large Language Models for Long-Context Understanding
25. NeurIPS 2025Return of ChebNet: Understanding and Improving an Overlooked GNN on Long Range Tasks
26. NeurIPS 2025Unleashing Hour-Scale Video Training for Long Video-Language Understanding
27. NeurIPS 2025VideoLucy: Deep Memory Backtracking for Long Video Understanding
28. NeurIPS 2024Hierarchical Visual Feature Aggregation for OCR-Free Document Understanding
29. NeurIPS 2024HourVideo: 1-Hour Video-Language Understanding
30. NeurIPS 2024LoTLIP: Improving Language-Image Pre-training for Long Text Understanding
31. NeurIPS 2024MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding
32. NeurIPS 2024MMLONGBENCH-DOC: Benchmarking Long-context Document Understanding with Visualizations
33. NeurIPS 2024Newswire: A Large-Scale Structured Database of a Century of Historical News
34. NeurIPS 2024SRFUND: A Multi-Granularity Hierarchical Structure Reconstruction Benchmark in Form Understanding
35. NeurIPS 2023Balance, Imbalance, and Rebalance: Understanding Robust Overfitting from a Minimax Game Perspective

常见问题

2023–2026 年四大会议一共有多少篇PDF 理解论文？

共 102 篇，按年份分别是 2023 年 6 篇、2024 年 22 篇、2025 年 35 篇、2026 年 39 篇。

NeurIPS 2026 有多少篇PDF 理解论文？

NeurIPS 2026 有 21 篇PDF 理解论文，上一届是 6 篇。

ICLR 2026 有多少篇PDF 理解论文？

ICLR 2026 有 6 篇PDF 理解论文，上一届是 8 篇。

ICML 2026 有多少篇PDF 理解论文？

ICML 2026 有 2 篇PDF 理解论文，上一届是 3 篇。

CVPR 2026 有多少篇PDF 理解论文？

CVPR 2026 有 10 篇PDF 理解论文，上一届是 18 篇。

哪个会议的PDF 理解论文最多？

CVPR，2023–2026 年共 39 篇。