Better color scaling in activation maps

app.py

@@ -36,8 +36,8 @@ def get_best_layout( num_imgs, img_size ):
     return n1_final, n2_final
 
 def create_plots( output_tensor, title, fontsize=24 ):
-    temp = output_tensor.detach().numpy()
-    mn, mx = temp.min(), temp.max()
+    # temp = output_tensor.detach().numpy()
+    # mn, mx = temp.min(), temp.max()
     img_list = [ output_tensor[0][n].detach().numpy() for n in range( output_tensor.shape[1] ) ]
     fig = plt.figure()
     M, N = get_best_layout( len( img_list ), img_list[0].shape )
@@ -47,7 +47,7 @@ def create_plots( output_tensor, title, fontsize=24 ):
         n = N0%N
         im = ax[m,n].imshow( img, cmap='gray' )
         ax[m,n].axis( 'off' )
-        im.set_clim( [ mn, mx ] )
+        # im.set_clim( [ mn, mx ] )
 
     plt.suptitle( title, fontsize=fontsize )
     # plt.subplots_adjust( hspace=0.01, wspace=0.01 )
@@ -81,11 +81,10 @@ model = learn.model.eval()
 # %% four-way-classifier.ipynb 5
 description='''
 A simple 4-way classifier that categorizes images as 'snake', 'bird', 'otter' or 'forest'. 
-Refined from an initial ResNet18 model downloaded from HuggingFace.
-The test images given here are chosen to demonstrate the effect of lack of training data on the classification outcome. 
+Refined from a pre-trained ResNet18 model downloaded from HuggingFace.
 
-The actual classification for each test image actually takes a very short time; the delay in predicing results here is due to 
-the extra step of platting the intermediate activation maps and inferred features in `matplotlib`. 
+The actual classification for each test image actually takes a very short time; the delay in displaying results is due to 
+the extra step of plotting the intermediate activation maps and inferred features in `matplotlib`. 
 
 
 **DISCLAIMER**: the images here are merely for demonstration purposes. I don't own any of them and I'm not making money from them.