Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
2
0
Stitching fragmented image features in the final encoder block unlocks global context in training-free semantic segmentation, boosting mIoU by 2 points without any additional training.