visual-grounding-reinforcement-learning

Improve LMM performance on high-resolution images by training models to ground reasoning on image regions through RL, learning spatial localization without requiring expensive grounding annotations.

Independent GitHub snapshot · Automated classification · Original documentation governs