@FuxiaoL
What if a VLM could teach itself from zero data? Meet MM-Zero: one base model that self-evolves by playing 3 roles: π§ Proposer β imagines visual concepts π» Coder β renders them in Python/SVG π Solver β reasons over the results No seed images. Just RL (GRPO) + execution feedback. @zli12321 @ChengsongH31219