debugging: xai output distortion

explanations.py

@@ -28,7 +28,23 @@ def preprocess_image(image, output_size=(300, 300)):
 
     return image_resized
 
-def show(img, output_size,p=False, **kwargs):
+
+def transform(image, original_size,output_size):
+    """
+    resize xai output back to original scale and pad to square-shape
+    """
+    h,w = original_size
+    image = cv2.resize(image,(h,w), interpolation = cv2.INTER_AREA)
+    if h > w:
+        padding = (h - w) // 2
+        image= cv2.copyMakeBorder(image, 0, 0, padding, padding, cv2.BORDER_CONSTANT, value=[0, 0, 0])
+    else:
+        padding = (w - h) // 2
+        image = cv2.copyMakeBorder(image, padding, padding, 0, 0, cv2.BORDER_CONSTANT, value=[0, 0, 0])
+    image = cv2.resize(image,output_size, interpolation = cv2.INTER_AREA)
+    return image
+
+def show(img, original_size, output_size,p=False, **kwargs):
 
     #img = preprocess_image(img, output_size=(output_size,output_size))
 
@@ -48,7 +64,8 @@ def show(img, output_size,p=False, **kwargs):
     # check if clip percentile
     if p is not False:
         img = np.clip(img, np.percentile(img, p), np.percentile(img, 100-p))
-    img = preprocess_image(img, output_size=(output_size,output_size))
+    
+    img = transform(img,original_size=original_size,output_size=output_size)
     plt.imshow(img, **kwargs)
     plt.axis('off')
     
@@ -56,7 +73,7 @@ def show(img, output_size,p=False, **kwargs):
     
 
 
-def explain(model, input_image,explain_method,nb_samples,size=600, n_classes=171) :
+def explain(model, input_image,h,w,explain_method,nb_samples,size=600, n_classes=171) :
     """
     Generate explanations for a given model and dataset.
     :param model: The model to explain.
@@ -130,8 +147,8 @@ def explain(model, input_image,explain_method,nb_samples,size=600, n_classes=171
                 phi = np.mean(phi, -1)
             #apply Gaussian smoothing
             phi_smoothed = cv2.GaussianBlur(phi, (5, 5), sigmaX=1.0, sigmaY=1.0)
-            show(X[0],output_size = size)
-            show(phi_smoothed, output_size = size,p=1, alpha=0.2)
+            show(X[0],original_size=(h,w),output_size = (size,size))
+            show(phi_smoothed, original_size=(h,w),output_size = (size,size),p=1, alpha=0.2)
             # show(X[0][:,size_repetitions:2*size_repetitions,:])
             # show(phi[:,size_repetitions:2*size_repetitions], p=1, alpha=0.4)
             plt.savefig(f'phi_{e}{i}.png')