Definition
Plain language
Pinpointing exactly when something happens in a video.
As stated in the literature
The video-understanding task of localizing a described event to a precise time interval; iterative zoom-in agents that re-request narrowing windows substantially outperform single-shot timestamp regression.
Why it matters: Pinpointing when events occur lets people search, clip, and reason about video instead of scrubbing through it manually.
For example, given the request 'find where the chef adds salt,' the system points to the exact 12-second stretch in a cooking video where it happens.
Heard on the show
“The result that genuinely surprised me wasn't even the headline — it was temporal grounding.”Episode 154 — How a 7B Model Out-Investigates a 72B One by Choosing What to Look At