default pass

src/oneat/NEATModels/neat_densevollnet.py

@@ -314,7 +314,7 @@ def predict(self,
            print(f'zero padded image shape ${self.image.shape}')
            self.second_pass_predict()
         if self.remove_markers == None:
-           self.image = create_sub_image(self.originalimage,self.config['imagez'],self.config['imagey'], self.config['imagex']) 
+           self.image = create_sub_image(self.originalimage,self.config['imagez'],self.config['imagey'], self.config['imagex'], self.stride) 
            self.default_pass_predict() 
 
 

src/oneat/NEATModels/neat_vollnet.py

@@ -312,7 +312,7 @@ def predict(self,
            self.second_pass_predict()
         if self.remove_markers == None:
 
-           self.image = create_sub_image(self.originalimage,self.config['imagez'],self.config['imagey'], self.config['imagex']) 
+           self.image = create_sub_image(self.originalimage,self.config['imagez'],self.config['imagey'], self.config['imagex'], self.stride) 
            self.default_pass_predict() 
 
 

src/oneat/NEATUtils/utils.py

@@ -398,21 +398,32 @@ def load_full_training_data(directory, filename, axes=None, verbose=True):
 
     return (X, Y), axes
 
-def create_sub_image(image, n, m, p):
-
+def create_sub_image(image, n, m, p, stride):
     t, z, y, x = image.shape
+    sub_images = []
 
-    z_remainder = z % n
-    y_remainder = y % m
-    x_remainder = x % p
-
-    new_z = z - z_remainder
-    new_y = y - y_remainder
-    new_x = x - x_remainder
-
-    sub_image = image[:, :new_z, :new_y, :new_x]
+    for t_idx in range(t):
+        sub_image_t = image[t_idx]
+
+        z_remainder = z % n
+        y_remainder = y % m
+        x_remainder = x % p
+
+        new_z = z - z_remainder + (n - 1)
+        new_y = y - y_remainder + (m - 1)
+        new_x = x - x_remainder + (p - 1)
+
+        sub_image = sub_image_t[:new_z, :new_y, :new_x]
+
+        pad_z = (new_z - z) % stride
+        pad_y = (new_y - y) % stride
+        pad_x = (new_x - x) % stride
+
+        sub_image_padded = np.pad(sub_image, ((0, pad_z), (0, pad_y), (0, pad_x)), mode='constant')
+
+        sub_images.append(sub_image_padded)
 
-    return sub_image
+    return np.array(sub_images)
 
 def pad_timelapse(image, pad_width):