A Simple Transformer-Based Model for Ego4D Natural Language Queries Challenge
This work addresses video grounding for ego-centric videos, an incremental improvement in a domain-specific challenge.
The paper tackled the Ego4D Natural Language Queries Challenge by developing a Transformer-based model for video grounding, achieving 12.64% Mean R@1 and ranking 2nd on the leaderboard, with R@5 scores up to 5.5 percentage points higher than the top solution.
This report describes Badgers@UW-Madison, our submission to the Ego4D Natural Language Queries (NLQ) Challenge. Our solution inherits the point-based event representation from our prior work on temporal action localization, and develops a Transformer-based model for video grounding. Further, our solution integrates several strong video features including SlowFast, Omnivore and EgoVLP. Without bells and whistles, our submission based on a single model achieves 12.64% Mean R@1 and is ranked 2nd on the public leaderboard. Meanwhile, our method garners 28.45% (18.03%) R@5 at tIoU=0.3 (0.5), surpassing the top-ranked solution by up to 5.5 absolute percentage points.