1
谷歌推出智能体视频理解功能：提升 Gemini 模型视频分析准确率，大幅降本减耗
IT之家 9 月 2 日消息，谷歌今日宣布在最新的 Gemini 3.7 Flash、3.6 Flash 和 3.5 Flash-Lite 模型中，推出 智能体视频理解（agentic video understanding） 功能。该新功能可提升视频分析准确率，同时大幅减少视频分析过程中的 Tok…
2
VideoGAIA: A Benchmark for General AI Assistants on Agentic Video Understanding
首个面向通用AI助手的智能体视频理解基准，测试AI在真实视频场景中的决策与交互能力。
arXiv:2608.14718v1 Announce Type: new Abstract: Video understanding is a fundamental task for evaluating the capabilities of multimodal large language…
3
Beyond Frame Selection: Rethinking Long-Video Understanding with MLLMs
跳出帧采样旧范式，重新思考多模态大模型如何真正理解长视频。
arXiv:2608.05592v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) have achieved strong progress in video understanding, yet it …
4
Do Video-LLMs Actually Watch? Diagnosing Character-Tracking Failures in Long-Form Video
诊断Video-LLM在长视频中跟踪角色的失败，揭示当前最强开源模型未能真正“观看”视频的局限。
arXiv:2607.11078v1 Announce Type: cross Abstract: Can a Video Large Language Model (Video-LLM) follow one person through a long video, keeping track o…
5
ECHO: Ego-Centric modeling of Human-Object interactions
第一视角怎么理解人-物交互？ECCV'26新方法用自我中心建模破题，值得视觉研究者跟进。
arXiv:2508.21556v3 Announce Type: replace Abstract: Modeling human-object interactions (HOI) from an egocentric perspective is a critical yet challeng…
6
UniVideo: Unified Understanding, Generation, and Editing for Videos
一个模型搞定视频理解、生成与编辑，UniVideo重新定义多模态AI边界。
arXiv:2510.08377v4 Announce Type: replace Abstract: Unified multimodal models have shown promising results in multimodal content generation and editin…
7
TimeThink: Reasoning with Time for Video LLMs
视频大模型只顾看画面？这项研究给LLM装上“时间思维”，让时序推理不再是短板。
arXiv:2607.05089v1 Announce Type: new Abstract: Video reasoning requires models to identify and verify temporally localized evidence within long video…
8
Agentic Very Long Video Understanding
提出智能体方法攻克超长视频理解难题，论文系统全面，含29页与8图表
arXiv:2601.18157v3 Announce Type: replace-cross Abstract: The advent of always-on personal AI assistants, enabled by all-day wearable devices such as …
9
Claude-real-video － any LLM can watch a video
开源工具让Claude等LLM真正“看懂”视频，不只抽帧，还结合音频分析动态内容。
Article URL: https://github.com/HUANGCHIHHUNGLeo/claude-real-video Comments URL: https://news.ycombinator.com/item?id=48766005 Points: 3 # Comments: 0
10
MotionAtlas: Detailed Region Captioning for Motion-Centric Videos
ECCV 2026新作，MotionAtlas攻克运动中心视频的精细区域描述，为视频理解提供更精准的细节维度。
arXiv:2606.29531v1 Announce Type: new Abstract: We propose MotionAtlas, a system for detailed captioning of motion-centric videos, comprising (1) a de…
11
AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression
一小时级音视频理解新突破，检索启发式token压缩让全模态大模型兼顾效率与精度。
arXiv:2606.24286v1 Announce Type: new Abstract: Multimodal Large Language Models have achieved remarkable progress in short-form audio-video understan…
12
EgoExo-Con: Exploring View-Invariant Video Temporal Understanding
第一视角与第三视角视频时间理解如何统一？这项研究提出EgoExo-Con框架，探索视角不变的时间建模，为多视角视频分析带来新思路。
arXiv:2510.26113v2 Announce Type: replace-cross Abstract: Do Video-LLMs have consistent temporal understanding when videos capture the same event from…
13
VideoAgent: All-in-One Framework for Video Understanding and Editing
一个统一框架搞定视频理解与编辑，专治长视频处理难题，AI视频工具的新方向。
arXiv:2606.23327v1 Announce Type: cross Abstract: Video editing has become essential in digital media creation, yet existing automated systems are res…
14
ViCoStream: Streaming VideoLLMs Can Run Beyond 100 FPS with Stage-Wise Coordinated Inference
流式视频大模型推理飙至100+FPS，ViCoStream用阶段协调机制破解实时处理瓶颈，视频理解应用提速必读。
arXiv:2606.19849v1 Announce Type: new Abstract: Streaming VideoLLMs must continuously process incoming video while maintaining low query latency, maki…
15
MVEB: Massive Video Embedding Benchmark
视频嵌入新基准MVEB：23项任务全面评估33个模型，揭示无模型独大现状
arXiv:2606.14958v1 Announce Type: cross Abstract: We introduce the Massive Video Embedding Benchmark (MVEB), a 23-task benchmark for video embeddings …
16
Scribby: A Multi-Level LLM Framework for Semantic Video Analysis
多级LLM框架Scribby，专攻长视频语义分析，为教育、直播等场景提供结构化解读
arXiv:2606.14762v1 Announce Type: cross Abstract: As video content continues to expand across educational platforms, recorded lectures, and live-strea…
17
3D-RFT: Reinforcement Fine-Tuning for Video-based 3D Scene Understanding
用强化学习微调视频输入，让3D场景理解精度与泛化性同步提升
arXiv:2603.04976v2 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards ( RLVR ) has emerged as a transformative para…
18
V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning
视频自监督学习新突破，V-JEPA 2.1巧妙解锁密集特征，大幅提升视频理解性能。
arXiv:2603.14482v3 Announce Type: replace Abstract: We present V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual re…
19
When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding
诊断视频理解中MLLM面对无正确答案时的表现缺陷，为提升模型鲁棒性提供关键洞见。
arXiv:2606.08239v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) have made substantial advancements in video understanding, …
20
Imagine Before You Predict: Interleaved Latent Visual Reasoning for Video Event Prediction
先想象再预测：论文提出交错潜在视觉推理新方法，提升视频事件预测准确性与可解释性。
arXiv:2606.05769v1 Announce Type: new Abstract: Video event prediction (VEP) requires models to infer unobserved future states from partial video evid…