stats_dataset = corda.CORDA(
encoder_df,
corda_base_path=corda_basepath,
rsna_base_path=rsna_basepath,
chest_base_path=chestxray_basepath,
transform=stats_transforms
)
stats_dataloader = torch.utils.data.DataLoader(
stats_dataset, batch_size=10,
shuffle=False, num_workers=10,
worker_init_fn=lambda id: utils.set_seed(seed),
pin_memory=True
)
mean, std = utils.get_mean_and_std(stats_dataloader)
print(f'Mean & std for corda+{args.pretrain}:', mean, std)
# CORDA ONLY (balance majority class)
train_df = corda_train_df
val_df = corda_val_df
test_df = corda_test_df
if args.train == 'corda':
train_df = pd.concat((
corda_train_df[corda_train_df.covid == 0].sample(n=84, random_state=42),
corda_train_df[corda_train_df.covid == 1].sample(n=84, random_state=42)
))
])
train_dataset = corda.CORDAChestXRay(stats_train_df,
chest_base_path=dataset_path,
corda_base_path=corda_basepath,
transform=transforms)
train_dataloader = torch.utils.data.DataLoader(
train_dataset,
batch_size=10,
shuffle=False,
num_workers=10,
worker_init_fn=lambda id: utils.set_seed(seed),
pin_memory=True)
mean, std = utils.get_mean_and_std(train_dataloader)
print(f'Mean & std for corda+chestxray (unprocessed):', mean, std)
# %%
train_transforms = torchvision.transforms.Compose([
torchvision.transforms.Resize(256),
torchvision.transforms.RandomAffine(0, translate=(0, 0.1),
scale=(1, 1.10)),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize(mean, std)
])
transforms = torchvision.transforms.Compose([
torchvision.transforms.Resize(256),
torchvision.transforms.CenterCrop(224),