analiza aiqbench
Event-Causal RAG addresses the critical bottleneck of maintaining coherent memory in ultra-long video streams by introducing a State-Event-State causal structure within a retrieval-augmented framework. This approach effectively mitigates context length limitations and event fragmentation, enabling accurate reasoning over videos exceeding one hour without prohibitive computational overhead. The accompanying ECV-1H benchmark provides a necessary standard for evaluating long-range temporal dependencies in future vision-language models.
Event-Causal RAG: A Retrieval-Augmented Generation Framework for Long Video Reasoning in Complex Scenarios
arXiv:2605.06185v2 Announce Type: replace Abstract: Large vision-language models perform well on short- and medium-length video understanding but still struggle to maintain coherent event memory and recover long-range relationships in ultra-long videos. End-to-end methods are limited by visual-token growth and context length, while fixed-segment retrieval often fragments complete events and weakens state-transition modeling.We propose Event-Causal RAG (EC-RAG), a lightweight retrieval-augmented framework for ultra-long and streaming video reasoning. A dual visual-audio sentinel mechanism segments video streams into semantically complete events, represented as State-Event-State (SES) structures that organize observable pre-event states, central events, and post-event states as event-local causal transitions. These transitions are stored in dual vector-graph memory and temporally connected through entity-consistent trajectories. During question answering, bidirectional graph retrieval recovers relevant predecessor and successor events, and answers are generated using both structured memory and the corresponding video evidence.We further introduce ECV-1H, an hour-scale long-video QA benchmark dedicated to directed event-causal reasoning, with all source videos exceeding one hour. It covers over 150 hours of untrimmed video and contains 1,251 fully human-annotated QA pairs. EC-RAG improves overall accuracy by 4.96\%--11.67\% across three open-source video foundation models and achieves consistent gains across public datasets. On a single RTX 5090 GPU with 32 GB of memory, EC-RAG can continuously process videos while maintaining controlled streaming memory usage.
Abstract:Large vision-language models perform well on short- and medium-length video understanding but still struggle to maintain coherent event memory and recover long-range relationships in ultra-long videos. End-to-end methods are limited by visual-token growth and context length, while fixed-segment retrieval often fragments complete events and weakens state-transition this http URL propose Event-Causal RAG (EC-RAG), a lightweight retrieval-augmented framework for ultra-long and streaming video reasoning. A dual visual-audio sentinel mechanism segments video streams into semantically complete events, represented as State-Event-State (SES) structures that organize observable pre-event states, central events, and post-event states as event-local causal transitions. These transitions are stored in dual vector-graph memory and temporally connected through entity-consistent trajectories. During question answering, bidirectional graph retrieval recovers relevant predecessor and successor events, and answers are generated using both structured memory and the corresponding video this http URL further introduce ECV-1H, an hour-scale long-video QA benchmark dedicated to directed event-causal reasoning, with all source videos exceeding one hour. It covers over 150 hours of untrimmed video and contains 1,251 fully human-annotated QA pairs. EC-RAG improves overall accuracy by 4.96\%--11.67\% across three open-source video foundation models and achieves consistent gains across public datasets. On a single RTX 5090 GPU with 32 GB of memory, EC-RAG can continuously process videos while maintaining controlled streaming memory usage.
Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2605.06185 [cs.AI]
(or arXiv:2605.06185v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2605.06185 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 309 cuvinte
Subiecte conexe
WorldScape Policy 2.0: Empowering Steerable World Action Modeling with Reasoning-Augmented Memory
arXiv · 22 iul. 2026
Amazon is investing in the Lean Focused Research Organization
Amazon Science · 26 iul. 2026
MemoryCard: Topic-Aware Multi-Modal Clue Compression for Long-Video Question Answering
arXiv · 5 iun. 2026
MemoryCard: Topic-Aware Multi-Modal Clue Compression for Long-Video Question Answering
arXiv · 31 aug. 2026