video-memory
epstein-files-cctv-video-memory
Epstein Files CCTV — Video Memory Index
Pre-built video memory index of DOJ Epstein Files Dataset 8 (MCC prison CCTV surveillance footage), created with av.
25 videos | ~24 hours of footage | 472 temporal event captions + 21 structured summaries + 21 analysis reports | searchable + queryable
Quickstart
Install av and download the pre-built database:
pip install pixelml-av
huggingface-cli download PixelML/epstein-files-cctv-video-memory av.db --local-dir .
Start querying… See the full description on the dataset page: https://huggingface.co/datasets/PixelML/epstein-files-cctv-video-memory.VideoMemory-Bench
VideoMemory-Bench
VideoMemory-Bench is a multiple-choice video question answering benchmark designed to evaluate video memory, long-range temporal understanding, cross-modal association, and reasoning over video content. The released benchmark split contains 5,360 QA pairs over 3,587 unique videos. Each QA item is attached to one video, one normalized level, one normalized category code, and one answer label.
This dataset card describes the Hugging Face release format. The… See the full description on the dataset page: https://huggingface.co/datasets/viktorcyt/VideoMemory-Bench.ebr-rag-ingest-no-cross-video-memory