fire
numpy
retry
tqdm
torch
torchvision
pillow
torch_optimizer
contrastive_learner>=0.1.0
axial-attention>=0.0.3
