Search papers, labs, and topics across Lattice.
Shenzhen Technology University
1
2
3
7
Current methods for evaluating visual-text compression are misleading because they conflate compression quality with the downstream model's ability to fill in the gaps.