DeltaStreamer: Towards Human-Like Agentic Streaming Video Understanding
DeltaStreamer combines change-aware memory and targeted replay for streaming video question answering.
DeltaStreamer is an agentic framework for streaming video understanding that records visual changes before a query arrives. A query-driven planner then selects required evidence, retrieves structured memory, and replays historical frames to recover details omitted during observation. Experiments report overall scores of 66.07% on OVO-Bench and 59.53% on StreamingBench, improving on representative streaming methods.
- DeltaStreamer records visual changes in structured memory before queries arrive.
- A planner retrieves memory and replays frames to recover omitted evidence.
- It scores 66.07% on OVO-Bench and 59.53% on StreamingBench.
- The design targets event transitions missed by compressed video histories.
From the original article
1 min read at huggingface.co
We propose DeltaStreamer, a novel agentic framework for streaming video understanding inspired by how humans observe, remember, and investigate. Streaming video understanding requires a system to retain evidence before knowing which details future queries will need. Existing approaches often obscure event transitions and omit relevant details when compressing video history, and cannot recover missing visual evidence.
By Boyuan Sun, Deshui Miao, Wenzhao Gao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.