MemLife: Curating and Reasoning over Long-Term Egocentric Video Memories
MemLife turns long first-person video histories into searchable text memories, then queries them without reopening the video.
The system builds entity-grounded, first-person episodes from egocentric video and uses a time-indexed agentic reader to find relevant memories. The paper says this avoids reprocessing hundreds of hours of raw clips for each query. MemLife beats the strongest training-free baseline by 4.6–12.0% across four long-horizon benchmarks, and MemOpt adds another 2.7–5.0% by training the memory writer for faithful, informative, retrievable entries. ArXiv · AI/CL/LG's note
The system builds entity-grounded, first-person episodes from egocentric video and uses a time-indexed agentic reader to find relevant memories. The paper says this avoids reprocessing hundreds of hours of raw clips for each query. MemLife beats the strongest training-free baseline by 4.6–12.0% across four long-horizon benchmarks, and MemOpt adds another 2.7–5.0% by training the memory writer for faithful, informative, retrievable entries. ArXiv · AI/CL/LG's note
score 5