- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
i am writing a code where i'm executing batch normalization using Tensorflow default methods (tf.nn.moments and tf.nn.batch_normalization).
When i try to compile the model for Movidius graph, the message "Unsupported Mean Operation" appears.
If i remove the batch normalization, everything works fine.
conv1 = tf.layers.conv2d(inputs=x, filters=32, kernel_size=[3,3], padding='same', use_bias=True, trainable=True)
mean1, variance1 = tf.nn.moments(conv1, axes=[0,1])
norm1 = tf.nn.batch_normalization(x= conv1, mean= mean1, variance= variance1, variance_epsilon=1e-5, offset=0, scale=0.95)
actv1 = tf.nn.relu(norm1)
The method moments don't have support for Movidius?
Thanks,
João
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@joaoreiser What is the network are you trying to compile? At the moment, the NCSDK doesn't have support for tf.nn.moments. It is a feature that we will be reviewing and considering for a future release.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel It is not an example one. It's a network that i coded it by myself. I will try the batch normalization another way. Any suggestion?
Thanks!

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page