[utils] Add load_real_mnist function

bob/learn/tensorflow/utils/util.py

@@ -17,6 +17,34 @@ def compute_euclidean_distance(x, y):
         d = tf.sqrt(tf.reduce_sum(tf.square(tf.subtract(x, y)), 1))
         return d
 
+def dense_to_one_hot(labels_dense, num_classes):
+  """
+  Convert class labels from scalars to one-hot vectors.
+
+  Taken from tensorflow/contrib/learn/python/learn/datasets/mnist.py
+  """
+  num_labels = labels_dense.shape[0]
+  index_offset = numpy.arange(num_labels) * num_classes
+  labels_one_hot = numpy.zeros((num_labels, num_classes))
+  labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
+  return labels_one_hot
+
+
+def load_real_mnist(one_hot=False, data_dir=None):
+    """
+    Return the original train and test data
+    """
+    import bob.db.mnist
+    db = bob.db.mnist.Database(data_dir=data_dir)
+
+    train_data, train_labels = db.data(groups=["train"])
+    test_data,  test_labels  = db.data(groups=["test"])
+
+    if one_hot:
+        train_labels = dense_to_one_hot(train_labels, 10)
+        test_labels = dense_to_one_hot(test_labels, 10)
+
+    return train_data, train_labels, test_data, test_labels
 
 def load_mnist(perc_train=0.9, data_dir=None):