EventMemAgent — online video QA with event-centric memory

Streams a video into a hierarchical memory (short-term frames + long-term captioned events), then runs a tool-using agent loop (search_memory, ocr, detect_objects) to answer your question.

Policy: lingcco/EventMemAgent-8B · paper · code

Video

Question

Long-term memory — captioned events

Short-term memory — recent frames

Examples

Example clips are excerpts from the Blender Foundation open movies Big Buck Bunny, Sintel and Tears of Steel — © Blender Foundation, CC BY 3.0 (bigbuckbunny.org, sintel.org, mango.blender.org).