add config

model_segvol_single.py

@@ -126,7 +126,6 @@ class SegVolProcessor():
         self.img_loader = transforms.LoadImage()
         self.transform4test = transforms.Compose(
             [
-                ForegroundNormalization(keys=["image"]),
                 DimTranspose(keys=["image", "label"]),
                 MinMaxNormalization(),
                 transforms.CropForegroundd(keys=["image", "label"], source_key="image"),
@@ -174,6 +173,7 @@ class SegVolProcessor():
         ct_voxel_ndarray = np.array(ct_voxel_ndarray).squeeze()
         ct_shape = ct_voxel_ndarray.shape
         ct_voxel_ndarray = np.expand_dims(ct_voxel_ndarray, axis=0)
+        ct_voxel_ndarray = self.ForegroundNorm(ct_voxel_ndarray)
         item['image'] = ct_voxel_ndarray
 
         # generate gt_voxel_ndarray
@@ -199,6 +199,19 @@ class SegVolProcessor():
         # transform
         return item['image'], item['label']
 
+    def ForegroundNorm(self, ct_narray):
+        ct_voxel_ndarray = ct_narray.copy()
+        ct_voxel_ndarray = ct_voxel_ndarray.flatten()
+        thred = np.mean(ct_voxel_ndarray)
+        voxel_filtered = ct_voxel_ndarray[(ct_voxel_ndarray > thred)]
+        upper_bound = np.percentile(voxel_filtered, 99.95)
+        lower_bound = np.percentile(voxel_filtered, 00.05)
+        mean = np.mean(voxel_filtered)
+        std = np.std(voxel_filtered)
+        ct_narray = np.clip(ct_narray, lower_bound, upper_bound)
+        ct_narray = (ct_narray - mean) / max(std, 1e-8)
+        return ct_narray
+
     def zoom_transform(self, ct_npy, gt_npy):
         item = {
             'image': ct_npy,
@@ -282,31 +295,6 @@ class DimTranspose(transforms.Transform):
             d[key] = np.swapaxes(d[key], -1, -3)
         return d
 
-class ForegroundNormalization(transforms.Transform):
-    def __init__(self, keys):
-        self.keys = keys
-    
-    def __call__(self, data):
-        d = dict(data)
-        
-        for key in self.keys:
-            d[key] = self.normalize(d[key])
-        return d
-    
-    def normalize(self, ct_narray):
-        ct_voxel_ndarray = ct_narray.copy()
-        ct_voxel_ndarray = ct_voxel_ndarray.flatten()
-        thred = np.mean(ct_voxel_ndarray)
-        voxel_filtered = ct_voxel_ndarray[(ct_voxel_ndarray > thred)]
-        upper_bound = np.percentile(voxel_filtered, 99.95)
-        lower_bound = np.percentile(voxel_filtered, 00.05)
-        mean = np.mean(voxel_filtered)
-        std = np.std(voxel_filtered)
-        ### transform ###
-        ct_narray = np.clip(ct_narray, lower_bound, upper_bound)
-        ct_narray = (ct_narray - mean) / max(std, 1e-8)
-        return ct_narray
-
 # prompts
 def generate_box(pred_pre, bbox_shift=None):
     meaning_post_label = pred_pre # [h, w, d]