Image Classification on ImageNet 1k (val) (Epoch Tracking)
68.1Accuracy @ Epoch 100SimSiam
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SimSiambatch size=256, negative pairs=false, momentum encoder=false, Backbone=ResNet-50, Pre-training views=two 224x224, Evaluation protocol=linear classification (single crop)2020.11 | 68.1 | 70 | 70.8 | 71.3 | |
| MoCo v2 (repro.+)batch size=256, negative pairs=true, momentum encoder=true, Backbone=ResNet-50, Pre-training views=two 224x224, Evaluation protocol=linear classification (single crop)2020.11 | 67.4 | 69.9 | 71 | 72.2 | |
| SimCLR (repro.+)batch size=4096, negative pairs=true, momentum encoder=false, Backbone=ResNet-50, Pre-training views=two 224x224, Evaluation protocol=linear classification (single crop)2020.11 | 66.5 | 68.3 | 69.8 | 70.4 | |
| BYOL (repro.)batch size=4096, negative pairs=false, momentum encoder=true, Backbone=ResNet-50, Pre-training views=two 224x224, Evaluation protocol=linear classification (single crop)2020.11 | 66.5 | 70.6 | 73.2 | 74.3 | |
| SwAV (repro.+)batch size=4096, negative pairs=false, momentum encoder=false, Backbone=ResNet-50, Pre-training views=two 224x224, Evaluation protocol=linear classification (single crop)2020.11 | 66.5 | 69.1 | 70.7 | 71.8 |