Strong but simple: A Baseline for Domain Generalized Dense Perception by CLIP-based Transfer Learning

Benchmark Model Rank Results
domain-generalization-on-gta-to-avgVLTSeg#6mIoU: 63.5
domain-generalization-on-gta5-to-cityscapesVLTSeg (EVA02-CLIP-L)#4mIoU: 65.6
robust-object-detection-on-dwdVLTDet#4mPC [AP50]: 36.9
semantic-segmentation-on-bdd100k-valVLTSeg#1mIoU: 72.5
semantic-segmentation-on-cityscapesVLTSeg#1Mean IoU (class): 86.4