Semantic Segmentation on CityScapes (mIoU, Accuracy)
74.6mIoUEVA-CLIP + UniRefiner
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EVA-CLIP + UniRefinerViT=8B/14, Evaluation Protocol=Linear probing2026.05 | 74.6 | 83.1 | |
| DINOv2 + UniRefinerViT=G/14, Evaluation Protocol=Linear probing2026.05 | 73.4 | 83 | |
| InternViT + UniRefinerViT=6B/14, Evaluation Protocol=Linear probing2026.05 | 71.8 | 80.9 | |
| DINOv2ViT=G/14, Evaluation Protocol=Linear probing2026.05 | 71.5 | 81.1 | |
| InternViTViT=6B/14, Evaluation Protocol=Linear probing2026.05 | 70.8 | 79.9 | |
| EVA-CLIPViT=8B/14, Evaluation Protocol=Linear probing2026.05 | 69.5 | 78.4 | |
| SigLIPv2 + UniRefinerViT=So/16, Evaluation Protocol=Linear probing2026.05 | 67.9 | 78.5 | |
| SigLIPv2ViT=So/16, Evaluation Protocol=Linear probing2026.05 | 63.4 | 74.6 | |
| CLIP + UniRefinerViT=G/14, Evaluation Protocol=Linear probing2026.05 | 61.7 | 71.5 | |
| CLIPViT=G/14, Evaluation Protocol=Linear probing2026.05 | 58.6 | 68.2 |