Table 3.
The parameter settings of SSRL
| Model | Backbone | Mask ratio | Lr | Lr_updated | Optimizer | Warm | Epoch | Batch |
|---|---|---|---|---|---|---|---|---|
| MAE | ViT-B/16 | 0.5 | 2.25e−4 | Cosine annealing | AdamW | 10 | 200 | 384 |
| SimCLR | ResNet-50 | - | 3e−3 | Cosine annealing | AdamW | 10 | 200 | 512 |
SSRL, self-supervised representation learning; MAE, masked autoencoder; SimCLR, simple framework for contrastive learning