[converter] Fix mobilenet conversion problem #2468
Conversation
dsmilkov
changed the title
pyu10055
left a comment
pyu10055
left a comment
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 1 approvals obtained (waiting on @dsmilkov, @nsthorat, and @pyu10055)
tfjs-converter/python/tensorflowjs/converters/tf_saved_model_conversion_v2.py, line 225 at r1 (raw file):
graph = tf.Graph() fuse_prelu.register_prelu_func(graph)
you can remove the register_prelu func as well
dsmilkov
left a comment
dsmilkov
left a comment
There was a problem hiding this comment.
Reviewable status:
tfjs-converter/python/tensorflowjs/converters/tf_saved_model_conversion_v2.py, line 225 at r1 (raw file):
Previously, pyu10055 (Ping Yu) wrote…
you can remove the register_prelu func as well
i can't unfortunately because it's used elsewhere, in between optimization passes
nsthorat
left a comment
nsthorat
left a comment
There was a problem hiding this comment.
Reviewed 2 of 2 files at r1.
Reviewable status:
dsmilkov
left a comment
dsmilkov
left a comment
There was a problem hiding this comment.
Reviewable status:
tfjs-converter/python/tensorflowjs/converters/tf_saved_model_conversion_v2.py, line 225 at r1 (raw file):
Previously, dsmilkov (Daniel Smilkov) wrote…
i can't unfortunately because it's used elsewhere, in between optimization passes
actually, turns out we can since we never reimport the graph after adding prelu
Fixes #2467
There is no need to re-import the graph after fusing just to extract the tensor values since there is a util method that can return
np.arrayfrom aTensorProto.BUG
This change is