研究人员正在开发先进的框架，以改进AI模型理解和推理长视频的方式。例如，Homer使用分层记忆系统，按时间因果联系组织信息，在M3-Bench-robot等基准测试中表现优于现有方法。Latent-VC通过在解码器中保留视觉记忆来解决“视觉锚定衰减”问题，从而实现更准确、更简洁的视频推理。EGAgent采用实体场景图和代理规划来实现以自我为中心的视频理解，而Light-Omni则提供了一种具有双重上下文状态以实现高效处理的反射式、轻量级方法。QSVideo专注于查询条件语义时间检索，通过提高相关性和多样性估计来增强VLM在长视频上的性能。
AI
arXiv:2607.02607v1 Announce Type: cross Abstract: Video reasoning requires Large Multimodal Models (LMMs) to remain grounded in dense evidence, yet existing systems largely adopt "read-once, generate-many" paradigm, in which visual grounding weakens during generation. This phenom…
arXiv:2601.18157v3 Announce Type: replace-cross Abstract: The advent of always-on personal AI assistants, enabled by all-day wearable devices such as smart glasses, demands a new level of contextual understanding, one that goes beyond short, isolated events to encompass the conti…
arXiv cs.AI
TIER_1English(EN)·Yixin Ji, Fanghua Ye, Juntao Li, Bo Zhao, Zexuan Qiu, Zhaopeng Tu, Liefeng Bo, Min Zhang·
arXiv:2607.02588v1 Announce Type: cross Abstract: Multimodal large language models excel on short clips but struggle on hour-long videos in an online setting, where frames are processed incrementally under limited memory. Existing online methods either retain compact visual repre…
Light-Omni is a multimodal agent framework that enables efficient video understanding through dual contextual states, achieving faster and more accurate video processing by eliminating iterative reasoning while maintaining semantic alignment.
Video-Oasis diagnostics reveal that half of existing video benchmarks can be solved without visual input, exposing significant capability gaps in current video understanding models.
arXiv:2607.05511v1 Announce Type: new Abstract: Agentic video understanding equips models with long-term memory to autonomously process and respond to continuous, long-horizon multimodal streams. However, advanced video agents often rely on ``detective-style'' iterative reasoning…
arXiv cs.CV
TIER_1English(EN)·Wei Ao, Lan Wang, Vishnu Naresh Boddeti·
arXiv:2607.04559v1 Announce Type: new Abstract: The performance of vision-language models (VLMs) in video understanding declines with increasing video duration, as video moments unrelated to the query confuse their language components. Multimodal retrieval has emerged as a critic…
