1. 2

Sign in to comment.

0
tldrbot6 hours ago

TL;DR of the paper, generated automatically.

Study on spotting mistakes in task videos as they happen, using instruction text.

  • Result: joint model scores 84.1% F1 vs. 80.7% for modular.
  • Slips: joint model gets 56.4% vs. 37.5%, the biggest gap.
  • Benchmark: EgoPER repartitioned so erroneous videos appear in every split.