Single Domain Generalization on CIFAR-10-C severity level 5 (test)
75.98Accuracy (Weather)L2D (Ours)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| L2D (Ours)Backbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing2021.08 | 75.98 | 69.16 | 73.29 | 72.02 | 72.61 | |
| M-ADABackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing2021.08 | 75.54 | 63.76 | 54.21 | 65.1 | 64.65 | |
| ME-ADABackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing, Implementation=Re-implemented by authors2021.08 | 74.44 | 71.37 | 66.47 | 70.83 | 70.77 | |
| ADABackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing, Implementation=Re-implemented by authors2021.08 | 72.67 | 67.04 | 39.97 | 66.62 | 61.58 | |
| d-SNEBackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing2021.08 | 67.9 | 56.59 | 33.97 | 61.83 | 55.07 | |
| CCSABackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing2021.08 | 67.66 | 57.81 | 28.73 | 61.96 | 54.04 | |
| ERMBackbone=WideResNet (16-4), Batch size=256, Corruption severity level=5, Optimizer=SGD with Nesterov momentum, Weight decay=0.0005, Learning rate schedule=cosine annealing2021.08 | 67.28 | 56.73 | 30.02 | 62.3 | 54.08 |