|
@@ -206,12 +206,12 @@ def resnet_v2(inputs,
|
|
|
if num_classes is not None:
|
|
|
net = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,
|
|
|
normalizer_fn=None, scope='logits')
|
|
|
+ logits = tf.squeeze(net, [1, 2], name='SpatialSqueeze')
|
|
|
# Convert end_points_collection into a dictionary of end_points.
|
|
|
end_points = slim.utils.convert_collection_to_dict(end_points_collection)
|
|
|
if num_classes is not None:
|
|
|
end_points['predictions'] = slim.softmax(net, scope='predictions')
|
|
|
- return net, end_points
|
|
|
-resnet_v2.default_image_size = 224
|
|
|
+ return logits, end_points
|
|
|
|
|
|
|
|
|
def resnet_v2_50(inputs,
|