spatial-ssrl-spatial-understanding-rl

【效率与协作】Improve spatial reasoning in vision-language models through five automatically-formulated pretext tasks (patch shuffling, flipping, inpainting, depth ordering, 3D position) that require zero human annotation, using only RGB/RGB-D images and RL optimization for 4% accuracy gains.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准