Search papers, labs, and topics across Lattice.
Shanghai AI Laboratory, Fudan University
1
0
3
Forget textual rules and coarse embeddings: a multimodal reward model that directly compares rendered visuals unlocks significant gains in vision-to-code RL.
