google-research
Data augument for supervised models
Hello, does the supervised comparison model you provided here use the same data enhancement method（random crop (with resize and random flip), random color distortion, and random Gaussian blur）?
No, they are trained with standard resnet settings (e.g. random crop augmenttion, no color augmentation or gaussian blur). With the extra augmentations and a longer training schedule, you could expect the supervised models to be improved roughly 0.5-2%.
Hello @chentingpc , does the supervised comparison model resnet50 1x use the same hyperparameter, optimizer and the scheduler as the example settings as follows? ( batch_size: 256, baselr: 0.1, weight_decay: 1e-4, optimizer: SGD(momentum: 0.99), scheduler: StepLR(step_size=30, gamma:0.1) Using the settings above, I can't reproduce your acc. 😢
