Merge pull request #27 from philipperemy/heatmap2

heatmap

README.md

@@ -101,7 +101,12 @@ dense_2/Softmax:0
 (128, 10)
 ```
 
-We can also visualise the activations. Here's another example using VGG16:
+We can visualise the activations. Here's another example using VGG16:
+
+```
+cd examples
+python vgg16.py
+```
 
 <p align="center">
  <img src="assets/cat.jpg">
@@ -114,6 +119,18 @@ We can also visualise the activations. Here's another example using VGG16:
  <br><i>Outputs of the first convolutional layer of VGG16.</i>
 </p>
 
+Also, we can visualise the heatmaps of the activations:
+
+```
+cd examples
+python heat_map.py
+```
+
+<p align="center">
+ <img src="assets/heatmap.png">
+</p>
+
+
 ### Repo views (since 2018/10/31)
 [![HitCount](http://hits.dwyl.io/philipperemy/keras-activations.svg)](http://hits.dwyl.io/philipperemy/keras-activations)
 

examples/heat_map.py

@@ -0,0 +1,32 @@
+from keras.applications.vgg16 import VGG16
+from keras.applications.vgg16 import decode_predictions
+from keras.applications.vgg16 import preprocess_input
+from keras.preprocessing.image import img_to_array
+
+model = VGG16()
+
+from PIL import Image
+import requests
+from io import BytesIO
+import numpy as np
+
+url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/1/14/Gatto_europeo4.jpg/250px-Gatto_europeo4.jpg'
+response = requests.get(url)
+image = Image.open(BytesIO(response.content))
+image = image.crop((0, 0, 224, 224))
+image = img_to_array(image)
+arr_image = np.array(image)
+image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
+image = preprocess_input(image)
+yhat = model.predict(image)
+label = decode_predictions(yhat)
+label = label[0][0]
+print('{} ({})'.format(label[1], label[2] * 100))
+
+import keract
+
+model.compile(optimizer='adam',
+ loss='categorical_crossentropy',
+ metrics=['accuracy'])
+activations = keract.get_activations(model, image)
+keract.display_heatmaps(activations, arr_image)

examples/vgg16.py

@@ -2,7 +2,6 @@
 from keras.applications.vgg16 import decode_predictions
 from keras.applications.vgg16 import preprocess_input
 from keras.preprocessing.image import img_to_array
-from keras.preprocessing.image import load_img
 
 model = VGG16()
 
@@ -13,10 +12,7 @@
 url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/1/14/Gatto_europeo4.jpg/250px-Gatto_europeo4.jpg'
 response = requests.get(url)
 image = Image.open(BytesIO(response.content))
-image = image.crop((0, 0, 244, 244))
-image.save('cat.jpg')
-
-image = load_img('cat.jpg', target_size=(224, 224))
+image = image.crop((0, 0, 224, 224))
 image = img_to_array(image)
 image = image.reshape((1, image.shape[0], image.shape[1], image.shape[2]))
 image = preprocess_input(image)
@@ -33,19 +29,3 @@
 activations = keract.get_activations(model, image)
 first = activations.get('block1_conv1/Relu:0')
 keract.display_activations(activations)
-#
-# import matplotlib.pyplot as plt
-#
-# fig = plt.figure(figsize=(12, 12))
-#
-# rows = 8
-# columns = 8
-#
-# first = activations.get('block1_conv1/Relu:0')
-#
-# for i in range(1, columns * rows + 1):
-# img = first[0, :, :, i - 1]
-# fig.add_subplot(rows, columns, i)
-# plt.imshow(img)
-# plt.axis('off')
-# plt.show()

keract/__init__.py

@@ -1,3 +1,3 @@
 from keract.keract import *
 
-__version__ = '2.3.0'
+__version__ = '2.4.0'

keract/keract.py

@@ -58,69 +58,71 @@ def get_activations(model, x, layer_name=None):
  return result
 
 
-def display_activations(activations,save=False):
+def display_activations(activations, save=False):
  import matplotlib.pyplot as plt
  max_rows = 8
  max_columns = 8
- for layer_name, first in activations.items():
- print(layer_name, first.shape, end=' ')
- if first.shape[0] != 1:
+ for layer_name, acts in activations.items():
+ print(layer_name, acts.shape, end=' ')
+ if acts.shape[0] != 1:
  print('-> Skipped. First dimension is not 1.')
  continue
- if len(first.shape) <= 2:
+ if len(acts.shape) <= 2:
  print('-> Skipped. 2D Activations.')
  continue
  print('')
  fig = plt.figure(figsize=(12, 12))
  plt.axis('off')
  plt.title(layer_name)
- for i in range(1, min(max_columns * max_rows + 1, first.shape[-1] + 1)):
- img = first[0, :, :, i - 1]
+ for i in range(1, min(max_columns * max_rows + 1, acts.shape[-1] + 1)):
+ img = acts[0, :, :, i - 1]
  fig.add_subplot(max_rows, max_columns, i)
  plt.imshow(img)
  plt.axis('off')
  if save:
- plt.savefig(layer_name.split("/")[0]+".png",bbox_inches="tight")
+ plt.savefig(layer_name.split('/')[0] + '.png', bbox_inches='tight')
  else:
  plt.show()
-def display_heatmaps(activations,image,save=False):
+
+
+def display_heatmaps(activations, image, save=False):
  from PIL import Image
  import matplotlib.pyplot as plt
  from sklearn.preprocessing import MinMaxScaler
  import numpy as np
  max_rows = 8
  max_columns = 8
- for layer_name, first in activations.items():
- print(layer_name, first.shape, end=' ')
- if first.shape[0] != 1:
+ for layer_name, acts in activations.items():
+ print(layer_name, acts.shape, end=' ')
+ if acts.shape[0] != 1:
  print('-> Skipped. First dimension is not 1.')
  continue
- if len(first.shape) <= 2:
+ if len(acts.shape) <= 2:
  print('-> Skipped. 2D Activations.')
  continue
  print('')
  fig = plt.figure(figsize=(12, 12))
  plt.axis('off')
  plt.title(layer_name)
- #computes values rquired to scale the activaions (which will form our heat map) to be in range 0-1
+ # computes values required to scale the activations (which will form our heat map) to be in range 0-1
  scaler = MinMaxScaler()
- scaler.fit(first.reshape(-1,1))
- for i in range(1, min(max_columns * max_rows + 1, first.shape[-1] + 1)):
- img = first[0, :, :, i - 1]
- #scales the activaions (which will form our heat map) to be in range 0-1
- img=scaler.transform(img)
- #resizes heatmap to be same dimesions of image
- img=Image.fromarray(img)
- img=img.resize((image.shape[0],image.shape[1]),Image.BILINEAR)
- img=np.array(img)
+ scaler.fit(acts.reshape(-1, 1))
+ for i in range(1, min(max_columns * max_rows + 1, acts.shape[-1] + 1)):
+ img = acts[0, :, :, i - 1]
+ # scale the activations (which will form our heat map) to be in range 0-1
+ img = scaler.transform(img)
+ # resize heatmap to be same dimensions of image
+ img = Image.fromarray(img)
+ img = img.resize((image.shape[0], image.shape[1]), Image.BILINEAR)
+ img = np.array(img)
  fig.add_subplot(max_rows, max_columns, i)
- #displays the image
- plt.imshow(image)
- #overlays a 70% tranparent heatmap onto the image
- #Lowest actiavtons are dark, highest are dark red, mid are yellow
- plt.imshow(img,alpha=0.3,cmap="jet",interpolation="bilinear")
+ # display the image
+ plt.imshow(image / 255.0)
+ # overlay a 70% transparent heat map onto the image
+ # Lowest activations are dark, highest are dark red, mid are yellow
+ plt.imshow(img, alpha=0.3, cmap='jet', interpolation='bilinear')
  plt.axis('off')
  if save:
- plt.savefig(layer_name.split("/")[0]+".png",bbox_inches="tight")
+ plt.savefig(layer_name.split('/')[0] + '.png', bbox_inches='tight')
  else:
  plt.show()

setup.py

@@ -2,7 +2,7 @@
 
 setup(
  name='keract',
- version='2.3.0',
+ version='2.4.0',
  description='Keras Activations and Gradients',
  author='Philippe Remy',
  license='MIT',