Search papers, labs, and topics across Lattice.
1
0
3
5
Slapping a learnable Gaussian kernel onto ViT self-attention boosts segmentation performance by over 6% on ADE20K, proving that locality still matters for vision.
