Deeplab code with LMS for both tensorflow and pytorch#181
Deeplab code with LMS for both tensorflow and pytorch#181naveenmiriyalu wants to merge 6 commits intoIBM:masterfrom
Conversation
smatzek
left a comment
There was a problem hiding this comment.
I have several comments.
- Which branch(es) it should merge to:
Based on the recreate steps text files I found it seems this code is intended to be run on top of WML CE 1.6.1
If that is correct, the readme:
examples/performance_models/deeplabv3/tensorflow-models/README.md
should be changed to reference the wmlce-1.6.1 branch rather than the powerai-1.6.0 branch.
I am trying to decide which branch this PR should be targeted for. I don't think it should be merged into master unless the intention is it can run on the latest version of WML CE and we plan to keep it alive to run on future versions of WML CE. It may be best to merge this into the wmlce-1.6.1 branch only.
Has this code been tested with WML CE 1.6.2 or future versions of WML CE?
What is the life expectancy of this code? Given the changes in TensorFlow 2, the DeepLabV3+ model will not work on TensorFlow 2. So this is another reason to not merge it into the master branch.
-
I think this deserves an overall readme in examples/performance_models explaining what the directory is for and what it contains. This should probably contain links to the recreate TXT files for both of the models.
-
What is the source of the PyTorch model? I didn't see that, but there are 215 files in this PR so I could have missed it. I think we should have a clear statement of where it is copied from and what the git commit hash was that was copied.
|
Created another pull request with the wmlce-1.6.1 branch |
4- Deeplabv3+ pytorch code is at the following location - https://github.com/naveenmiriyalu/powerai/tree/deeplab_code/examples/performance_models/pytorch-deeplab-xception