Elad Eban
Elad Eban
+ @ayp-google @mengdong Thanks for rising this issue. We routinely work with ResNet and other complicated models so I don't think that the complexity is the issue. Are you using...
took another look. it might be `reduce_mean` in line 542. can you apply the regularizer to take `inputs` prior to that? qq: are you using `channels_first`?
`channels_first` is the problem. We assume `channels_last`... Note that you need to use `channels_last` only during structure leanring, later you could revert back to (faster?) `channel_first`.
there should not be any problem with a colab. Do you see a valid value for the cost and regularization strength? is it the same values you get on the...
Dear not a big fish, I am sorry to say that for now MorphNet does NOT support conv3d. What you have observed is the `op_regularizer_manager` that looking for an op...
I believe the the README contains some information. Try adding the GammaFLOPs regularizer loss to your training, and see what happens.
@Aspirinkb Interesting point :) Could you suggest a specific model (and open source code for it) that **you** will be happy to see the MorphNet demo for.
Sorry about the delay, but we don't have the bandwidth to do these examples. We will be more than happy to show case such examples here if somewhat is willing...