Keypoint Matching on SPair-71K
63.91Accuracy (d=0)DINOv2
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DINOv2Arch=ViT-L/14, Pre-training Data=LVD-142M, Note=distilled from DINOv2-G, Image Resolution=800 x 8002025.07 | 63.91 | 53.11 | 51.91 | 56.92 | |
| DINOv2Arch=ViT-B/14, Pre-training Data=LVD-142M, Note=distilled from DINOv2-G, Image Resolution=800 x 8002025.07 | 63.04 | 51.09 | 48.76 | 55.98 | |
| FrancaArch=ViT-L/14, Pre-training Data=LAION-600M, Image Resolution=800 x 8002025.07 | 58.5 | 46.74 | 48.28 | 51.76 | |
| DINOv2Arch=ViT-L/14, Pre-training Data=IN-21K, Note=reproduced on IN-21K without distillation, Image Resolution=800 x 8002025.07 | 57.6 | 42.91 | 44.17 | 50.68 | |
| FrancaArch=ViT-B/14, Pre-training Data=IN-21K, Image Resolution=800 x 8002025.07 | 53.46 | 41.9 | 44.2 | 45.6 | |
| DINOv2Arch=ViT-B/14, Pre-training Data=IN-21K, Note=reproduced on IN-21K without distillation, Image Resolution=800 x 8002025.07 | 42.82 | 33.38 | 34.82 | 38.71 | |
| OpenCLIPArch=ViT-L/14, Pre-training Data=LAION-2B, Image Resolution=800 x 8002025.07 | 22.73 | 20.18 | 20.37 | 21.26 | |
| OpenCLIPArch=ViT-B/14, Pre-training Data=LAION-2B, Image Resolution=800 x 8002025.07 | 18.31 | 16.78 | 17.05 | 17.63 | |
| SigLIP2Arch=ViT-B/14, Pre-training Data=WebLI, Image Resolution=800 x 8002025.07 | 9.29 | 5.96 | 5.82 | 7.83 | |
| SigLIP2Arch=ViT-L/14, Pre-training Data=WebLI, Image Resolution=800 x 8002025.07 | 8.66 | 5.38 | 5.05 | 6.88 |