Update app.py

app.py

@@ -7,7 +7,7 @@ from alibi.explainers import IntegratedGradients
 from alibi.datasets import load_cats
 from alibi.utils.visualization import visualize_image_attr
 import numpy as np
-from PIL import Image
+from PIL import Image, ImageFilter
 import io
 import time
 import os
@@ -46,34 +46,42 @@ def do_process(img, baseline):
   lstPreds = decode_predictions(preds, top=3)[0]
   dctPreds = {lstPreds[i][1]: round(float(lstPreds[i][2]),2) for i in range(len(lstPreds))}
   predictions = preds.argmax(axis=1)
-  if baseline is 'white':
+  if baseline == 'white':
     baselines = bls = np.ones(instance.shape).astype(instance.dtype)
-  elif baseline is 'black':
+    img_flt = Image.fromarray(np.uint8(np.squeeze(baselines)*255))
+  elif baseline == 'black':
     baselines = bls = np.zeros(instance.shape).astype(instance.dtype)
+    img_flt = Image.fromarray(np.uint8(np.squeeze(baselines)*255))
+  elif baseline == 'blur':
+    img_flt = img.filter(ImageFilter.GaussianBlur(5))
+    baselines = image.img_to_array(img_flt)
+    baselines = np.expand_dims(baselines, axis=0)
+    baselines = preprocess_input(baselines)
   else:
     baselines = np.random.random_sample(instance.shape).astype(instance.dtype)
+    img_flt = Image.fromarray(np.uint8(np.squeeze(baselines)*255))
   explanation = ig.explain(instance,
                          baselines=baselines,
                          target=predictions)
   attrs = explanation.attributions[0]
-  fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(10, 10))
   fig, ax = visualize_image_attr(attr=attrs.squeeze(), original_image=img, method='blended_heat_map',
-                      sign='all', show_colorbar=True, title=None,
-                      plt_fig_axis=(fig, ax), use_pyplot=False)
+                      sign='all', show_colorbar=True, title=baseline,
+                      plt_fig_axis=None, use_pyplot=False)
   fig.tight_layout()
   buf = io.BytesIO()
   fig.savefig(buf)
   buf.seek(0)
   img_res = Image.open(buf)
-  return img_res, dctPreds
- 
+  return img_res, img_flt, dctPreds
+  
 input_im = gr.inputs.Image(shape=(224, 224), image_mode='RGB', 
                          invert_colors=False, source="upload",
                          type="pil")
 input_drop = gr.inputs.Dropdown(label='Baseline (default: random)',
-	                  choices=['black', 'white', 'random'], default='random', type='value')
+	                  choices=['random', 'black', 'white', 'blur'], default='random', type='value')
 
 output_img = gr.outputs.Image(label='Output of Integrated Gradients', type='pil')
+output_base = gr.outputs.Image(label='Baseline image', type='pil')
 output_label = gr.outputs.Label(label='Classification results', num_top_classes=3)
 
 title = "XAI - Integrated gradients"
@@ -83,7 +91,7 @@ article="<p style='text-align: center'><a href='https://github.com/mawady/colab-
 iface = gr.Interface(
     fn=do_process, 
     inputs=[input_im, input_drop], 
-    outputs=[output_img,output_label],
+    outputs=[output_img,output_base,output_label]
     live=False,
     interpretation=None,
     title=title,