Comment 'transforms.CropForegroundd'

Pre-training with 'transforms.CropForegroundd' was crashing to zero. Details: #12
ivadomed · Apr 6, 2024 · cbcf7ea · cbcf7ea
1 parent bb7cd2c
commit cbcf7ea
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/pretraining_and_finetuning/transforms.py b/pretraining_and_finetuning/transforms.py
@@ -12,7 +12,7 @@ def train_transforms(crop_size, patch_size, task="pretraining"):
             transforms.LoadImaged(keys=all_keys, image_only=False),
             transforms.EnsureChannelFirstd(keys=all_keys),
             transforms.Orientationd(keys=all_keys, axcodes="RPI"),
-            transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),   # crop everything outside the SC mask
+            # transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),   # crop everything outside the SC mask
             # NOTE: spine interpolation with order=2 is spline, order=1 is linear
             transforms.Spacingd(keys=all_keys, pixdim=(0.8, 0.8, 0.8), mode=(2, 1)),
             transforms.ResizeWithPadOrCropd(keys=all_keys, spatial_size=crop_size),
@@ -60,7 +60,7 @@ def train_transforms(crop_size, patch_size, task="pretraining"):
             transforms.LoadImaged(keys=all_keys),
             transforms.EnsureChannelFirstd(keys=all_keys),
             transforms.Orientationd(keys=all_keys, axcodes="RPI"),
-            transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),   # crop everything outside the SC mask
+            # transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),   # crop everything outside the SC mask
             # NOTE: spine interpolation with order=2 is spline, order=1 is linear
             transforms.Spacingd(keys=all_keys, pixdim=(0.8, 0.8, 0.8), mode=(2, 1, 1)),
             transforms.ResizeWithPadOrCropd(keys=all_keys, spatial_size=crop_size),
@@ -106,7 +106,7 @@ def val_transforms(crop_size, task="pretraining"):
             transforms.LoadImaged(keys=all_keys, image_only=False),
             transforms.EnsureChannelFirstd(keys=all_keys),
             transforms.Orientationd(keys=all_keys, axcodes="RPI"),
-            transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),
+            # transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),
             transforms.Spacingd(keys=all_keys, pixdim=(0.8, 0.8, 0.8), mode=(2, 1)),
             transforms.ResizeWithPadOrCropd(keys=all_keys, spatial_size=crop_size),
             transforms.DivisiblePadd(keys=all_keys, k=2**5),   # pad inputs to ensure divisibility by no. of layers nnUNet has (5)
@@ -120,7 +120,7 @@ def val_transforms(crop_size, task="pretraining"):
             transforms.LoadImaged(keys=all_keys, image_only=False),
             transforms.EnsureChannelFirstd(keys=all_keys),
             transforms.Orientationd(keys=all_keys, axcodes="RPI"),
-            transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),
+            # transforms.CropForegroundd(keys=all_keys, source_key="label_sc"),
             transforms.Spacingd(keys=all_keys, pixdim=(0.8, 0.8, 0.8), mode=(2, 1, 1)),
             transforms.ResizeWithPadOrCropd(keys=all_keys, spatial_size=crop_size),
             transforms.DivisiblePadd(keys=all_keys, k=2**5),   # pad inputs to ensure divisibility by no. of layers nnUNet has (5)