Search papers, labs, and topics across Lattice.
1
0
3
Counterintuitively, VLMs can achieve higher VQA accuracy by intentionally degrading visual inputs, suggesting that high-resolution details can act as noise that hinders reasoning.
