Decoupled superpixel clustering + ViT for near real-time semantic segmentation: 53.77% mIoU at 54.8 ms/image on Cityscapes with a frozen ResNet-50 and parameter-free SLIC (Georgia Tech CS 7641, Team 45)
machine-learning computer-vision pytorch semantic-segmentation superpixels slic cityscapes georgia-tech cs7641 vision-transformer
-
Updated
Jul 21, 2026 - Python