| """ |
| Validate an already trained cropping UNet on a test split, with the training augmentation applied. |
| |
| Same purpose as val_trained_unet.py in the repository root, which validates without augmentation. |
| Paths not given on the command line come from the site config (see config.example.yaml): |
| |
| python scripts/validation/validate_crop_unet.py --unet-path <unet_best.pth> |
| """ |
|
|
| import argparse |
|
|
| import torchio as tio |
|
|
| from auto_detect_breast_mri.config import resolve_path |
| from auto_detect_breast_mri.data.transforms import ImageToTensor, RandomCropOrPad, ZNormalization, masking_method |
| from auto_detect_breast_mri.models.unet_crop import UNet_Crop |
| from auto_detect_breast_mri.data.loaders import get_subjects_dataloader |
| from auto_detect_breast_mri.data.metadata import get_uka_metatensor |
| def build_parser(): |
| parser = argparse.ArgumentParser(description=__doc__, |
| formatter_class=argparse.RawDescriptionHelpFormatter) |
| parser.add_argument("--unet-path", required=True, |
| help="pth file of the trained cropping UNet") |
| parser.add_argument("--data-path", default=None, |
| help="image root. Config key: data_root") |
| parser.add_argument("--feature-path", default=None, |
| help="metadata export. Config key: metadata_file") |
| parser.add_argument("--split-files-folder", default=None, |
| help="folder holding the split files. Config key: split_root") |
| parser.add_argument("-b", "--batch-size", type=int, default=4) |
| parser.add_argument("-c", "--fold", type=int, default=0, help="outer fold whose test set is used") |
| parser.add_argument("-n", "--number-of-epochs", type=int, default=10, |
| help="how often the validation set is passed through. Default: 10") |
| parser.add_argument("--device", default="cuda", choices=["cuda", "cpu"]) |
| return parser |
|
|
|
|
| if __name__ == '__main__': |
| args = build_parser().parse_args() |
|
|
| path_base = resolve_path(args.data_path, "data_root", "image root") |
| feature_path = resolve_path(args.feature_path, "metadata_file", "metadata export") |
| split_root = resolve_path(args.split_files_folder, "split_root", "folder with the split files") |
| test_split = f"{split_root.rstrip('/')}/fold{args.fold}/stratified_test_set" |
|
|
| protocol = ['Dyn_0'] |
| pre_image_shape = (32, 512, 512) |
| augmentation = tio.Compose([ |
| tio.RandomFlip((0, 1, 2), flip_probability=0.5), |
| RandomCropOrPad(pre_image_shape), |
| ZNormalization(per_channel=True, percentiles=(0.5, 99.5), masking_method=masking_method), |
| tio.RandomNoise(std=(0.25, 0.5)), |
| ImageToTensor() |
| ]) |
|
|
| |
| in_channel = len(protocol) |
| CropNet = UNet_Crop(path_to_unet=args.unet_path, input_channel=in_channel, running_device=args.device, |
| show_validation=True, criterion='dice') |
|
|
| |
| feature_dataframe = get_uka_metatensor(0, feature_path) |
| test_loader = get_subjects_dataloader(path_base, feature_dataframe, pre_image_shape, |
| transform=augmentation, protocol=protocol, |
| data_selection_file_sceleton=test_split, |
| batch_size=args.batch_size, fold=args.fold, |
| segmentation_task=True) |
| CropNet.validate(args.number_of_epochs, test_loader) |
|
|
