Add support to print the predictions and to use mean_value when mean_file is not supplied

python/caffe/io.py

@@ -256,7 +256,14 @@ def set_mean(self, in_, mean):
             if len(ms) != 3:
                 raise ValueError('Mean shape invalid')
             if ms != self.inputs[in_][1:]:
-                raise ValueError('Mean shape incompatible with input shape.')
+                print(self.inputs[in_])
+    		in_shape = self.inputs[in_][1:]
+    		m_min, m_max = mean.min(), mean.max()
+    		normal_mean = (mean - m_min) / (m_max - m_min)
+    		mean = resize_image(normal_mean.transpose((1,2,0)),
+                        in_shape[1:]).transpose((2,0,1)) * \
+                        (m_max - m_min) + m_min
+                #raise ValueError('Mean shape incompatible with input shape.')
         self.mean[in_] = mean
 
     def set_input_scale(self, in_, scale):

python/classify.py

@@ -5,6 +5,7 @@
 By default it configures and runs the Caffe reference ImageNet model.
 """
 import numpy as np
+import pandas as pd
 import os
 import sys
 import argparse
@@ -80,6 +81,17 @@ def main(argv):
         help="Order to permute input channels. The default converts " +
              "RGB -> BGR since BGR is the Caffe default by way of OpenCV."
     )
+    parser.add_argument(
+        "--labels_file",
+        default=os.path.join(pycaffe_dir,
+                "../data/ilsvrc12/synset_words.txt"),
+        help="Readable label definition file."
+    )
+    parser.add_argument(
+        "--print_results",
+        action='store_true',
+        help="Write output text to stdout rather than serializing to a file."
+    )
     parser.add_argument(
         "--ext",
         default='jpg',
@@ -93,6 +105,10 @@ def main(argv):
     mean, channel_swap = None, None
     if args.mean_file:
         mean = np.load(args.mean_file)
+    else: 
+        # channel-wise mean 
+        mean = np.array([104,117,123])
+
     if args.channel_swap:
         channel_swap = [int(s) for s in args.channel_swap.split(',')]
 
@@ -126,12 +142,48 @@ def main(argv):
 
     # Classify.
     start = time.time()
-    predictions = classifier.predict(inputs, not args.center_only)
+    scores = classifier.predict(inputs, not args.center_only).flatten()
     print("Done in %.2f s." % (time.time() - start))
 
+#   The script has been updated to support --print_results option. 
+#   Ref - http://stackoverflow.com/questions/37265197/classify-py-is-not-taking-argument-print-results
+#   However, the labels format supported here has been modified, such that the file can have shorttext 
+#   corresponding to the category classes instead of the general format. 
+#   The commented part correspond to the general format of labels file which has mapping between 
+#   synset_id and the text. 
+
+    if args.print_results:
+#        with open(args.labels_file) as f:
+#          labels_df = pd.DataFrame([
+#               {
+#                   'synset_id': l.strip().split(' ')[0],
+#                   'name': ' '.join(l.strip().split(' ')[1:]).split(',')[0]
+#               }
+#               for l in f.readlines()
+#            ])
+#        labels_df.synset_id = labels_df.synset_id.astype(np.int64)
+#        labels = labels_df.sort('synset_id')['name'].values
+
+        labels_file = open(args.labels_file, 'r')
+        labels = labels_file.readlines()
+
+        indices = (-scores).argsort()[:5]
+#        predictions = labels[indices]
+#        print(predictions)
+#        meta = [
+#                   (p, '%.5f' % scores[i])
+#                   for i, p in zip(indices, predictions)
+#               ]
+#        print meta
+        print("---------------------------------")    
+        print("The top 5 predictions are")
+        for i in indices:
+           print('%.4f %s' % (scores[i] , labels[i].strip('\n')))
+        print("---------------------------------")
+
     # Save
     print("Saving results into %s" % args.output_file)
-    np.save(args.output_file, predictions)
+    np.save(args.output_file, scores)
 
 
 if __name__ == '__main__':