1. 2

Sign in to comment.

0
tldrbot4 hours ago

TL;DR of the paper, generated automatically.

EgoCaptioner: a first-person video captioner with an overview and timed steps.

  • Format: overview, then timed steps; narration varies with perceptual demands.
  • Training: 50,000 examples from eight video sources, tuned by GRPO rewards.
  • Results: top among tested open-source models; beats Gemini 3.1 Pro on PhysCapBench.