center_crop

training_scripts/unigradICON/ITK_transforms.py

@@ -36,12 +36,33 @@ def itk_crop_foreground(image: itk.Image, monai_crop_pad:monai.transforms.CropFo
 
     return image
 
-def itk_resize_with_pad_or_crop(image: itk.image, pad_or_crop:monai.transforms.ResizeWithPadOrCrop):
+def itk_resize_with_pad_or_crop(image: itk.image, pad_or_crop:monai.transforms.SpatialCrop):
     torch_image = torch.tensor(itk.GetArrayFromImage(image))
+
+    boundary_coords = pad_or_crop.com
 
 
-def itk_crop(image:itk.image, crop:monai.transforms.SpatialPad):
-    pass
+def itk_crop(image:itk.image, cropper:monai.transforms.CenterSpatialCrop):
+
+    torch_image = torch.tensor(itk.GetArrayFromImage(image))
+    crop = cropper.compute_slices(spatial_size=torch_image.shape[:])
+
+    lower = [s.start for s in reversed(crop)]
+    upper = [dim_len - s.stop for dim_len, s in reversed(list(zip(torch_image.shape, crop)))]
+
+    print(lower, upper)
+
+
+    filter = itk.CropImageFilter[type(image), type(image)].New()
+    filter.SetInput(image)
+    filter.SetLowerBoundaryCropSize(lower)
+    filter.SetUpperBoundaryCropSize(upper)
+
+    filter.Update()
+
+    image = filter.GetOutput()
+
+    return image
 
 
 

training_scripts/unigradICON/test_ITK_transforms.py

@@ -4,13 +4,13 @@
 import tqdm
 import torch
 
-from monai.transforms import CropForeground, SpatialPad, ResizeWithPadOrCrop
+from monai.transforms import CropForeground, SpatialPad, ResizeWithPadOrCrop, SpatialCrop
 with open(f"/playpen-raid2/lin.tian/projects/icon_lung/ICON/training_scripts/brain_t1_pipeline/splits/train.txt") as f:
     image_paths = f.readlines()
 f_path = image_paths[0].split(".nii.gz")[0] + "_restore_brain.nii.gz"
 import ITK_transforms
 import importlib
-def doit():
+def test_itk_crop_foreground():
     importlib.reload(ITK_transforms)
 
     # crop foreground test
@@ -29,6 +29,47 @@ def doit():
 
     print(torch.sum((cropped_monai - cropped_itk)**2))
 
+def test_itk_crop_filter():
+
+    importlib.reload(ITK_transforms)
+
+    image = itk.imread(f_path)
+
+    transform = ResizeWithPadOrCrop([175, 175, 175]).cropper
+
+    print(transform)
+
+    image_t = torch.tensor(np.asarray(image))[None]
+
+    import pdb
+#    pdb.set_trace()
+    cropped_monai = transform(image_t)
+
+    print("Cropped monai shape", cropped_monai.shape)
+
+    crop = transform.compute_slices(spatial_size=image_t.shape[1:])
+
+    print(dir(crop[0]))
+
+
+
+
+    cropped_itk =torch.tensor( np.asarray(ITK_transforms.itk_crop(image, transform)))[None, None]
+
+    print("itk, monai:", cropped_itk.shape, cropped_monai.shape)
+
+
+    print(torch.sum((cropped_monai - cropped_itk)**2))
+
+def test_itk_pad_image_filter():
+    importlib.reload(ITK_transforms)
+
+    image = itk.imread(f_path)
+
+    image = ITK_transforms.itk_crop_foreground(image, CropForeground())
+
+    tranform = SpatialPad([175, 175, 175])
+