Skip to content

Fix #447: Changed format for exporting models to MetaGraphDef for Tensorflow#455

Open
abhigyan7 wants to merge 9 commits intoCloud-CV:masterfrom
abhigyan7:tf_metagraphdef
Open

Fix #447: Changed format for exporting models to MetaGraphDef for Tensorflow#455
abhigyan7 wants to merge 9 commits intoCloud-CV:masterfrom
abhigyan7:tf_metagraphdef

Conversation

@abhigyan7
Copy link
Copy Markdown
Contributor

This PR makes Fabrik export models as MetaGraphDef instead of Graphdef, which fixes #447. Doing this allows models built in Fabrik to be imported into Tensorflow seamlessly.

@Ram81
Copy link
Copy Markdown
Member

Ram81 commented Nov 16, 2018

@haabhi nice work, can you test whether all models in FabrikZoo which were working before are still working after the change or not.

@yashdusing
Copy link
Copy Markdown
Contributor

@haabhi are you going to make the changes in import_graphdef as well ?

@abhigyan7
Copy link
Copy Markdown
Contributor Author

@yashdusing Yes, I am doing that at the moment. I am working on importing both metagraphdefs and graphdefs so the model zoo models and models exported from fabrik both work.

@abhigyan7
Copy link
Copy Markdown
Contributor Author

@Ram81 @yashdusing
Now both binary and text metagraphdefs and graphdefs can be used to import tensorflow models in to fabrik. I have tested this with the prototxt models in the example/tensorflow directory and some random .meta and .pb files.

Comment thread ide/tasks.py
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Models exported as graphdef protos for Tensorflow cannot be trained.

3 participants