[SOLVED]The mxnet model cannot be transformed


#1

When I use “from_mxnet” transformation, the following error is reported, and the model can be loaded normally when “mxnet” is used, so the model should be no problem.

mxnet.base.MXNetError: Cannot find argument 'layout', Possible Arguments:

thanks a lot!


#2

Could you reveal which model you’re importing?


#3

my model is mobilenet1.0 Trained through mxnet example,the file is /mxnet/exmple/gluon/image_classification.py
The command to run is

python3 image_classification.py --model mobilenet1.0 --gpus 0 --mode hybrid --dtype float16

In order to obtain .json add “ net.export(“mobilenet”, epoch=1)” in line 205

float32 also have this problem.


#4

solve it ! Because the mxnet version of the training model is inconsistent with the running version.