Image Classification on ImageNet-1K (Linear Probing/k-NN Evaluation)
82.6Linear Probing AccuracyUniFlow
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| UniFlowSize=ViT-L, Evaluation Protocol=linear probing2025.10 | 82.6 | — | — | — | |
| iBOTArch.=ViT-L/16, Par.=307M, im/s=102, Epo.=200, pre-training=ImageNet-22K2021.11 | 82.3 | 72.9 | — | — | |
| iBOTArch.=ViT-L/16, Par.=307M, im/s=102, Epo.=12002021.11 | 81 | 78 | — | — | |
| SCLRv2Arch.=RN152w3, Par.=794M, im/s=46, Epo.=2000, selective kernel=true2021.11 | 79.8 | 73.1 | — | — | |
| BYOLArch.=RN200w2, Par.=250M, im/s=123, Epo.=20002021.11 | 79.6 | 73.9 | — | — | |
| iBOTArch.=ViT-B/16, Par.=85M, im/s=312, Epo.=16002021.11 | 79.5 | 77.1 | — | — | |
| iBOTArch.=Swin-T/14, Par.=28M, im/s=593, Epo.=12002021.11 | 79.3 | 76.2 | — | — | |
| MAGESize=ViT-L, Evaluation Protocol=linear probing2025.10 | 78.9 | — | — | — | |
| EsViTArch.=Swin-T/14, Par.=28M, im/s=593, Epo.=12002021.11 | 78.7 | 77 | — | — | |
| iBOTArch.=Swin-T/7, Par.=28M, im/s=726, Epo.=12002021.11 | 78.6 | 75.3 | — | — | |
| DINOArch.=ViT-B/16, Par.=85M, im/s=312, Epo.=16002021.11 | 78.2 | 76.1 | — | — | |
| EsViTArch.=Swin-T/7, Par.=28M, im/s=726, Epo.=12002021.11 | 78.1 | 75.7 | — | — | |
| MoCo v3Size=ViT-H, Evaluation Protocol=linear probing2025.10 | 78.1 | — | — | — | |
| iBOTArch.=ViT-S/16, Par.=21M, im/s=1007, Epo.=32002021.11 | 77.9 | 75.2 | — | — | |
| DINOArch.=ViT-S/16, Par.=21M, im/s=1007, Epo.=32002021.11 | 77 | 74.5 | — | — | |
| MoCov3Arch.=ViT-B/16, Par.=85M, im/s=312, Epo.=12002021.11 | 76.7 | — | — | — | |
| SupervisedEp=1002023.10 | 76.5 | — | — | — | |
| MAESize=ViT-L, Evaluation Protocol=linear probing2025.10 | 75.8 | — | — | — | |
| SwAVArch.=RN50, Par.=23M, im/s=1237, Epo.=24002021.11 | 75.3 | 65.7 | — | — | |
| DINOArch.=RN50, Par.=23M, im/s=1237, Epo.=32002021.11 | 75.3 | 67.5 | — | — | |
| SwAV (mc)Ep=8002023.10 | 75.3 | — | 66.3 | — | |
| DINO (mc)Ep=8002023.10 | 75.3 | — | 67.5 | — | |
| CARP (mc)Ep=4002023.10 | 75.3 | — | 67.7 | — | |
| MoCov3Arch.=RN50, Par.=23M, im/s=1237, Epo.=16002021.11 | 74.6 | — | — | — | |
| SwAV (mc)Ep=4002023.10 | 74.6 | — | 65 | — | |
| MoCo-v3Ep=10002023.10 | 74.6 | — | 68.9 | — | |
| BYOLEp=10002023.10 | 74.3 | — | 66.6 | — | |
| CARP (mc)Ep=2002023.10 | 74.2 | — | 66.5 | — | |
| SwAV (mc)Ep=2002023.10 | 73.9 | — | 63.7 | — | |
| SwAVArch.=ViT-S/16, Par.=21M, im/s=1007, Epo.=24002021.11 | 73.5 | 66.3 | — | — | |
| BEiTSize=ViT-L, Evaluation Protocol=linear probing2025.10 | 73.5 | — | — | — | |
| MoCov3Arch.=ViT-S/16, Par.=21M, im/s=1007, Epo.=12002021.11 | 73.4 | — | — | — | |
| CARPEp=4002023.10 | 73 | — | 67.6 | — | |
| CARP (mc)Ep=1002023.10 | 72.5 | — | 63.5 | — | |
| SwAV (mc)Ep=1002023.10 | 72.1 | — | 61.8 | — | |
| VFMTokSize=ViT-L, Evaluation Protocol=linear probing2025.10 | 69.4 | — | — | — | |
| EPProbed on=ImageNet-1K, Backbone=MAE ViT-B, Configuration=EP322025.06 | — | — | — | 70.5 | |
| EPProbed on=StanfordCars, Backbone=MAE ViT-B, Configuration=EP322025.06 | — | — | — | 51.7 | |
| EPProbed on=Food-101, Backbone=MAE ViT-B, Configuration=EP322025.06 | — | — | — | 58.2 | |
| frozen MAE backboneBackbone=MAE ViT-B2025.06 | — | — | — | 46.1 | |
| LoRATuned on=ImageNet-1K, Backbone=MAE ViT-B, Rank=rho=82025.06 | — | — | — | 72.3 | |
| LoRATuned on=StanfordCars, Backbone=MAE ViT-B, Rank=rho=82025.06 | — | — | — | 45.8 | |
| LoRATuned on=Food-101, Backbone=MAE ViT-B, Rank=rho=82025.06 | — | — | — | 56.7 |