Input channel layout with layout optimization #16924
Unanswered
francis2tm
asked this question in
General
Replies: 1 comment
-
Hmm I'm not sure if there is any current functionality to remove an op in an ONNX graph -- you could try some third-party libraries like ONNX modifier or ONNX tool |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello,
So, right now, I have a model that receives images which are channel last (NHWC). The ONNX default is NCHW. Therefore, my model needs to have a transpose op in the beginning:
However, in the docs you say that after the model is optimized, the "graph uses NCHWc layout instead of NCHW layout".
So... The pretty expensive transpose OP added in the beginning just becomes useless, correct?
I'd like very much to remove that transpose OP. Any ideas?
Beta Was this translation helpful? Give feedback.
All reactions