google-research / google-research/morph-net
Batch Norm layers Assertion Error
Open
- Dominant language
- Python
- Stars
- 1k
- Forks
- 151
- PR merge metrics
- No merged PRs in 30d
Description
I'm getting an assertion error in tpu_util.py when I use batchnorm layers. I noticed that there are two 'ops' associated with the batchnorm layers:
- 'model/conv1_bn_td/cond/ReadVariableOp_2'
- 'model/conv1_bn_td/cond/ReadVariableOp_2/Switch'
The error happens when the following line is run on the ''model/conv1_bn_td/cond/ReadVariableOp_2/Switch' op:
`assert len(op.inputs) == 1`
There are two items in this inputlist. I can simply comment out this line and everything seems to work, but am I doing anything destructive if I do comment this line out?
Contributor guide
Assessment
This issue has not been assessed yet.