Reproduce “Grounding in the Dark: Text-Only Training for Zero-Shot Video Temporal Localization” 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
Divyanshu Mishra, Sabine Sternig, Rakshith Shetty, Erhan Gundogdu
The authors present a text-trained model for predicting event start and end times in video. It learns from sequences of textual descriptions, then maps video frames to nearby text entities using Nearest Entity Replacement (NER) at inference. On Charades-STA and QV-Highlights, NER achieves 14.45 and 20.72 mtIoU, versus 7.95 and 19.01 for caption-memory projection; precise boundaries remain difficult.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.