1. 2

Sign in to comment.

0
tldrbot2 hours ago

TL;DR of the paper, generated automatically.

A data engine that trains vision-language models to zoom in on image details.

  • Approach: fine-tunes on automated data, skipping extra RL complexity.
  • Results: beats the SFT+RL state-of-the-art using 3.1x less data.
  • Simple RL: beats prior complex setups by 9.59 points on average.