Added a GPU check for NVIDIA in Makefile

This fix will only work on Linux. We would need to use 'ioreg' for Mac

Makefile

@@ -32,7 +32,11 @@ endif
 	$(PIP) install --upgrade -r $(REQUIREMENTS)
 
 	@echo -e $(TAG)Installing Eyeballer$(END)
-	$(PIP) install --upgrade --editable .
+ifeq ($(shell lspci | grep NVIDIA),NVIDIA))
+	$(PIP) install --upgrade --editable .[tf_gpu]
+else
+	$(PIP) install --upgrade --editable .[tf]
+endif
 
 ifeq ($(strip $(MAKECMDGOALS)),conda)
 	@echo -e $(TAG)Activate virtualenv to run eyeballer: $(END)

eyeballer/eyeballer.py

@@ -65,8 +65,10 @@ def pretty_print_evaluation(results):
     for label in DATA_LABELS:
         print(f"{label} Precision Score: {results[label]['precision']:4.2%}")
         print(f"{label} Recall Score: {results[label]['recall']:4.2%}")
+    print(f"'None of the above' Precision: {results['none_of_the_above_precision']:4.2%}")
+    print(f"'None of the above' Recall: {results['none_of_the_above_recall']:4.2%}")
+    print(f"All or nothing Accuracy: {results['all_or_nothing_accuracy']:4.2%}")
     print(f"Overall Binary Accuracy: {results['total_binary_accuracy']:4.2%}")
-    print(f"All or nothing accuracy: {results['all_or_nothing_accuracy']:4.2%}")
     print(f"Top 10 worst predictions: {results['top_10_worst'][1]}")
 
 

eyeballer/model.py

@@ -1,5 +1,6 @@
 import os
 import random
+import sys
 
 # Prevent Tkinter Dependency
 import matplotlib
@@ -50,7 +51,7 @@ class EyeballModel:
     image_width, image_height = 224, 224
     input_shape = (image_width, image_height, 3)
 
-    def __init__(self, print_summary=False, weights_file="weights.h5", seed=None):
+    def __init__(self, weights_file, print_summary=False, seed=None):
         """Constructor for model class.
 
         Keyword arguments:
@@ -95,10 +96,16 @@ class EyeballModel:
         self.training_labels = self.training_labels.sample(frac=1)
 
         if weights_file is not None and os.path.isfile(weights_file):
-            self.model.load_weights(weights_file)
+            try:
+                self.model.load_weights(weights_file)
+            except OSError:
+                print(f"ERROR: Unable to open weights file '{weights_file}'")
+                sys.exit(-1)
             print("Loaded model from file.")
         else:
-            print("No model loaded from file")
+            if weights_file is not None:
+                raise FileNotFoundError
+            print("WARN: No model loaded from file. Generating random model")
 
         # Data augmentation
         augmentor = Augmentor.Pipeline()
@@ -306,8 +313,38 @@ class EyeballModel:
         stats["all_or_nothing_accuracy"] = accuracy_score(ground_truth, predictions)
         stats["top_10_best"] = self._top_images(evaluation_generator, predictions, best=True)
         stats["top_10_worst"] = self._top_images(evaluation_generator, predictions, best=False)
+        stats["none_of_the_above_recall"] = self._none_of_the_above_recall(ground_truth, predictions)
+        stats["none_of_the_above_precision"] = self._none_of_the_above_precision(ground_truth, predictions)
         return stats
 
+    def _none_of_the_above_recall(self, labels, predictions):
+        """Returns the recall score for the 'none of the above' images.
+        That means, all the images that don't have a category.
+        """
+        total_count = 0
+        correct_count = 0
+        for item in zip(labels.astype(int), predictions.astype(int)):
+            # Is this a none of the above?
+            if not item[0].any():
+                total_count += 1
+                if not item[1].any():
+                    correct_count += 1
+        return correct_count / total_count
+
+    def _none_of_the_above_precision(self, labels, predictions):
+        """Returns the precision score for the 'none of the above' images.
+        That means, all the images that don't have a category.
+        """
+        total_count = 0
+        correct_count = 0
+        for item in zip(labels.astype(int), predictions.astype(int)):
+            # Is this a none of the above prediction?
+            if not item[1].any():
+                total_count += 1
+                if not item[0].any():
+                    correct_count += 1
+        return correct_count / total_count
+
     def _top_images(self, labels, predictions, top_k=10, best=False):
         """Collect top-k best or top-k worst predicted images
 

setup.py

@@ -6,7 +6,7 @@ setup(
     packages=find_packages(exclude=["tests"]),
     url='https://github.com/BishopFox/eyeballer',
     license='GNU General Public License v3.0',
-    author='Daniel Petro',
+    author='Bishop Fox',
     author_email='',
     description='',
     entry_points={
@@ -22,9 +22,12 @@ setup(
         'pandas',
         'pillow',
         'sklearn',
-        'tensorflow',
         'keras @ git+https://github.com/keras-team/keras',
     ],
+    extras_require={
+    'tf': ['tensorflow'],
+    'tf_gpu': ['tensorflow-gpu'],
+    },
     dependency_links=[
     ],
 )

tests/PredictTest.py

@@ -2,7 +2,8 @@ import errno
 import os
 import unittest
 
-from eyeballer.model import EyeballModel, DATA_LABELS
+from eyeballer.model import EyeballModel
+
 
 class PredictTest(unittest.TestCase):
     def setUp(self):
@@ -10,9 +11,9 @@ class PredictTest(unittest.TestCase):
         if not os.path.isfile(weights_file):
             print("Error: Symlink the latest weights file to 'latest_weights.h5'")
             raise FileNotFoundError(
-            errno.ENOENT,
-            os.strerror(errno.ENOENT),
-            weights_file)
+                errno.ENOENT,
+                os.strerror(errno.ENOENT),
+                weights_file)
 
         model_kwargs = {
             "weights_file": weights_file,
@@ -51,15 +52,15 @@ class PredictTest(unittest.TestCase):
 
         self.assertEquals(results_one, results_two)
 
-    def get_prediction_by_name(self,results):
+    def get_prediction_by_name(self, results):
         prediction = None
         max_val = 0.0
 
         print(results)
         for k, v in results[0].items():
             if not isinstance(v, str) and v > max_val:
-                    prediction = k
-                    max_val = v
+                prediction = k
+                max_val = v
         return prediction
 
     def test_predict_custom404(self):