Image Classification on iNaturalist 2021
87.3Top-1 AccuracyDINOv2 (dst)
Evaluation Results
| Method | Links | |
|---|---|---|
| DINOv2 (dst)Arch=ViT-L/14, Data=LVD, Ep.=-, Probe=X-Blk2026.03 | 87.3 | |
| FrancaArch=ViT-L/14, Data=LAION, Ep.=(3.2), Probe=X-Blk2026.03 | 86.2 | |
| DINOv2 (dst)Arch=ViT-B/14, Data=LVD, Ep.=-, Probe=X-Blk2026.03 | 84.3 | |
| DINOv2 (dst)Arch=ViT-L/14, Data=LVD, Ep.=-, Probe=CLS2026.03 | 83.2 | |
| H8(ViT-H/16)FLOPs=77.5 x 10^9, Evaluation Protocol=linear2025.05 | 82.2 | |
| I8(ViT-H/16)FLOPs=77.7 x 10^9, Evaluation Protocol=linear2025.05 | 81.9 | |
| ViT-H/16FLOPs=127.7 x 10^9, Evaluation Protocol=linear2025.05 | 81.7 | |
| H8(ViT-H/16)FLOPs=77.5 x 10^9, Evaluation Protocol=k-NN2025.05 | 81.4 | |
| H8(ViT-L/16)FLOPs=37.7 x 10^9, Evaluation Protocol=linear2025.05 | 81.3 | |
| I8(ViT-L/16)FLOPs=38.5 x 10^9, Evaluation Protocol=linear2025.05 | 81.2 | |
| FrancaArch=ViT-B/14, Data=IN-22k, Ep.=(90), Probe=X-Blk2026.03 | 81 | |
| ViT-H/16FLOPs=127.7 x 10^9, Evaluation Protocol=k-NN2025.05 | 81 | |
| I8(ViT-H/16)FLOPs=77.7 x 10^9, Evaluation Protocol=k-NN2025.05 | 80.9 | |
| ViT-L/16FLOPs=61.9 x 10^9, Evaluation Protocol=linear2025.05 | 80.9 | |
| H8(ViT-L/16)FLOPs=37.7 x 10^9, Evaluation Protocol=k-NN2025.05 | 80.8 | |
| CAPIArch=ViT-L/14, Data=IN-1k, Ep.=6394, Probe=X-Blk2026.03 | 80.7 | |
| ViT-L/16FLOPs=61.9 x 10^9, Evaluation Protocol=k-NN2025.05 | 80.5 | |
| DINOv2 (dst)Arch=ViT-B/14, Data=LVD, Ep.=-, Probe=CLS2026.03 | 80.4 | |
| I8(ViT-L/16)FLOPs=38.5 x 10^9, Evaluation Protocol=k-NN2025.05 | 80.4 | |
| DINOv2 (dst)Arch=ViT-S/14, Data=LVD, Ep.=-, Probe=X-Blk2026.03 | 78.9 | |
| BootlegArch=ViT-L/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 77.1 | |
| FrancaArch=ViT-L/14, Data=LAION, Ep.=(3.2), Probe=CLS2026.03 | 77 | |
| MAEArch=ViT-L/16, Data=IN-1k, Ep.=1600, Probe=X-Blk2026.03 | 76.3 | |
| DINOv2 (dst)Arch=ViT-L/14, Data=LVD, Ep.=-, Probe=Patch2026.03 | 76.3 | |
| CrossMAEArch=ViT-L/16, Data=IN-1k, Ep.=800, Probe=X-Blk2026.03 | 75.1 | |
| BootlegArch=ViT-B/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 74.2 | |
| DINOv2 (dst)Arch=ViT-S/14, Data=LVD, Ep.=-, Probe=CLS2026.03 | 74 | |
| FrancaArch=ViT-B/14, Data=IN-22k, Ep.=(90), Probe=CLS2026.03 | 73.6 | |
| data2vec 2.0Arch=ViT-L/16, Data=IN-1k, Ep.=200, Probe=X-Blk2026.03 | 72.7 | |
| FrancaArch=ViT-L/14, Data=LAION, Ep.=(3.2), Probe=Patch2026.03 | 71.5 | |
| DINOv2 (dst)Arch=ViT-B/14, Data=LVD, Ep.=-, Probe=Patch2026.03 | 71.4 | |
| MAEArch=ViT-B/16, Data=IN-1k, Ep.=1600, Probe=X-Blk2026.03 | 70.7 | |
| CrossMAEArch=ViT-B/16, Data=IN-1k, Ep.=800, Probe=X-Blk2026.03 | 70.2 | |
| Fine-Tunedbackbone=DINOv3 ViT-B/162025.12 | 68.4 | |
| BootlegArch=ViT-S/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 67.4 | |
| DINOv2 (dst)Arch=ViT-S/14, Data=LVD, Ep.=-, Probe=Patch2026.03 | 64.2 | |
| FrancaArch=ViT-B/14, Data=IN-22k, Ep.=(90), Probe=Patch2026.03 | 63.2 | |
| I-JEPAArch=ViT-B/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 63 | |
| data2vec 2.0Arch=ViT-B/16, Data=IN-1k, Ep.=200, Probe=X-Blk2026.03 | 61.3 | |
| I-JEPAArch=ViT-L/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 61.3 | |
| BootlegArch=ViT-L/16, Data=IN-1k, Ep.=600, Probe=CLS2026.03 | 61.2 | |
| Linear Probebackbone=DINOv3 ViT-B/162025.12 | 60.9 | |
| CrossMAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=X-Blk2026.03 | 59.8 | |
| Task Matrixbackbone=DINOv3 ViT-B/16, best layer=112025.12 | 59.1 | |
| BootlegArch=ViT-B/16, Data=IN-1k, Ep.=600, Probe=CLS2026.03 | 58.3 | |
| MAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=X-Blk2026.03 | 57.5 | |
| BootlegArch=ViT-L/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 53.2 | |
| MAEArch=ViT-L/16, Data=IN-1k, Ep.=1600, Probe=CLS2026.03 | 51.8 | |
| BootlegArch=ViT-B/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 50.9 | |
| CrossMAEArch=ViT-L/16, Data=IN-1k, Ep.=800, Probe=CLS2026.03 | 50.1 | |
| I-JEPAArch=ViT-S/16, Data=IN-1k, Ep.=600, Probe=X-Blk2026.03 | 48.4 | |
| BootlegArch=ViT-S/16, Data=IN-1k, Ep.=600, Probe=CLS2026.03 | 47.8 | |
| CAPIArch=ViT-L/14, Data=IN-1k, Ep.=6394, Probe=Patch2026.03 | 47.6 | |
| data2vec 2.0Arch=ViT-S/16, Data=IN-1k, Ep.=200, Probe=X-Blk2026.03 | 47.4 | |
| MAEArch=ViT-L/16, Data=IN-1k, Ep.=1600, Probe=Patch2026.03 | 43.9 | |
| MAEArch=ViT-B/16, Data=IN-1k, Ep.=1600, Probe=CLS2026.03 | 43.3 | |
| CrossMAEArch=ViT-L/16, Data=IN-1k, Ep.=800, Probe=Patch2026.03 | 43.2 | |
| BootlegArch=ViT-S/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 42.6 | |
| CrossMAEArch=ViT-B/16, Data=IN-1k, Ep.=800, Probe=CLS2026.03 | 41.7 | |
| data2vec 2.0Arch=ViT-L/16, Data=IN-1k, Ep.=200, Probe=CLS2026.03 | 41.5 | |
| I-JEPAArch=ViT-B/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 41.4 | |
| I-JEPAArch=ViT-L/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 41.1 | |
| data2vec 2.0Arch=ViT-L/16, Data=IN-1k, Ep.=200, Probe=Patch2026.03 | 38.6 | |
| CrossMAEArch=ViT-B/16, Data=IN-1k, Ep.=800, Probe=Patch2026.03 | 38.3 | |
| MAEArch=ViT-B/16, Data=IN-1k, Ep.=1600, Probe=Patch2026.03 | 37.4 | |
| data2vec 2.0Arch=ViT-B/16, Data=IN-1k, Ep.=200, Probe=CLS2026.03 | 31.3 | |
| data2vec 2.0Arch=ViT-B/16, Data=IN-1k, Ep.=200, Probe=Patch2026.03 | 29.7 | |
| I-JEPAArch=ViT-S/16, Data=IN-1k, Ep.=600, Probe=Patch2026.03 | 26.8 | |
| CrossMAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=CLS2026.03 | 26.7 | |
| MAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=CLS2026.03 | 26.2 | |
| CrossMAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=Patch2026.03 | 25.7 | |
| MAEArch=ViT-S/16, Data=IN-1k, Ep.=800, Probe=Patch2026.03 | 22.3 | |
| dino.txtResolution=336, Pre-training Dataset=LVTD-2.3B, Evaluation Protocol=Zero-shot2024.12 | 19.4 | |
| dino.txtResolution=224, Pre-training Dataset=LVTD-2.3B, Evaluation Protocol=Zero-shot2024.12 | 18.8 | |
| DFNResolution=224, Pre-training Dataset=DFN-2B, Evaluation Protocol=Zero-shot2024.12 | 17.6 | |
| SigLIPResolution=384, Pre-training Dataset=WebLi, Evaluation Protocol=Zero-shot2024.12 | 17.5 | |
| data2vec 2.0Arch=ViT-S/16, Data=IN-1k, Ep.=200, Probe=Patch2026.03 | 15.4 | |
| SigLIPResolution=256, Pre-training Dataset=WebLi, Evaluation Protocol=Zero-shot2024.12 | 14.8 | |
| data2vec 2.0Arch=ViT-S/16, Data=IN-1k, Ep.=200, Probe=CLS2026.03 | 14.6 | |
| MetaCLIPResolution=224, Pre-training Dataset=MetaCLIP-2.5B, Evaluation Protocol=Zero-shot2024.12 | 10.2 | |
| EVA-02-CLIPResolution=336, Pre-training Dataset=Merged-2B, Evaluation Protocol=Zero-shot2024.12 | 6.4 | |
| CLIPResolution=336, Pre-training Dataset=WIT-400M, Evaluation Protocol=Zero-shot2024.12 | 5.7 |