Image Classification on ImageNet ILSVRC 2012 (val) (Epoch-Specific Accuracy)
69.6Top-1 Acc (100 ep)GroCo
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GroCoBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=1024, Views=4x2242023.01 | 69.6 | 70.6 | 71.3 | |
| WhiteningBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=4096, Views=4x2242023.01 | 69.4 | — | 72.6 | |
| GroCoBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=1024, Views=2x2242023.01 | 69.2 | 70.4 | 71.1 | |
| Barlow TwinsBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=2048, Views=2x2242023.01 | 68.7 | — | — | |
| VICRegBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=2048, Views=2x2242023.01 | 68.6 | — | — | |
| SimSiamBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=256, Views=2x2242023.01 | 68.1 | 70 | 70.8 | |
| MoCo v2+Backbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=256, Views=2x2242023.01 | 67.4 | 69.9 | 71 | |
| SimCLR+Backbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=4096, Views=2x2242023.01 | 66.5 | 68.3 | 69.8 | |
| SwAV+Backbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=4096, Views=2x2242023.01 | 66.5 | 69.1 | 70.7 | |
| BYOL+Backbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=4096, Views=2x2242023.01 | 66.5 | 70.6 | 73.2 | |
| Max-MarginBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=256, Views=2x2242023.01 | 63.8 | — | — | |
| GroCoBackbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=1024, Views=4x224, k=202023.01 | 61.8 | 63.6 | 64.8 | |
| GroCoBackbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=1024, Views=2x224, k=202023.01 | 60.5 | 62.9 | 64 | |
| SimSiamBackbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=256, Views=2x224, k=202023.01 | 57.4 | — | — | |
| SimCLR+Backbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=4096, Views=2x224, k=202023.01 | 53.8 | 57.2 | 59.2 | |
| MoCo v2Backbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=256, Views=2x224, k=202023.01 | — | 55.6 | — | |
| ReSSLBackbone=ResNet-50, Evaluation Protocol=Linear Probing, Batch Size=256, Views=2x2242023.01 | — | 69.9 | — | |
| SwAVBackbone=ResNet-50, Evaluation Protocol=k-NN, Batch Size=4096, Views=2x224, k=202023.01 | — | — | 61.3 |