"mask_func":
mask_func,
"seed":
1,
"filter": [filter_acquisition_no_fs],
"num_sym_slices":
0,
"multi_slice_gt":
False,
"keep_mask_as_func":
True,
"transform":
torchvision.transforms.Compose([
CropOrPadAndResimulate(n),
Flatten(0, -3),
Normalize(reduction="mean", use_target=True),
], ),
}
test_data = AlmostFixedMaskDataset
test_data = test_data("val", **test_data_params)
vols = range(30)
slices_in_vols = [test_data.get_slices_in_volume(vol_idx) for vol_idx in vols]
slices_selected = [
range((lo + hi) // 2, (lo + hi) // 2 + 1) for lo, hi in slices_in_vols
]
samples = np.concatenate(slices_selected)
X_0 = torch.stack([test_data[s][2] for s in samples])
X_0 = to_complex(X_0.to(device))
"lr": 1e-4,
"eps": 1e-4,
"weight_decay": 1e-5
}],
"scheduler":
torch.optim.lr_scheduler.StepLR,
"scheduler_params": {
"step_size": 1,
"gamma": 1.0
},
"acc_steps": [1],
"train_transform":
torchvision.transforms.Compose([
CropOrPadAndResimulate((368, 368)),
Flatten(0, -3),
Normalize(reduction="mean", use_target=False),
Jitter(1e1, 0.0, 1.0),
]),
"val_transform":
torchvision.transforms.Compose([
CropOrPadAndResimulate((368, 368)),
Flatten(0, -3),
Normalize(reduction="mean", use_target=False),
], ),
"train_loader_params": {
"shuffle": True,
"num_workers": 8
},
"val_loader_params": {
"shuffle": False,
"num_workers": 8