Reproduce “Retrieval-Augmented Online Textual Memory for Episodic Memory Video Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Raffaele Cali, Giuseppe Lando, Rosario Forte, Antonino Furnari
The Parent retrieval method searches small caption subchunks, then gives the reasoner the full captions they came from, while holding the generated video memory fixed. On QAEgo4D-Closed, it reaches 56.4% answer accuracy with 3,600 average query-time input tokens, versus 56.0% and 10,880 for a full-memory system; accuracy peaks at K=12, then declines despite rising temporal coverage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.