How to set environment variable TF_Keras = 1 for onnx conversion?

问题

Recently updated to tensorflow 2.0 and am having trouble getting my .h5 models into .onnx . Used to be a very simple procedure but now I am having an issue. When I run the following code:

# onnx testing
import onnx    
import keras2onnx
import os
import tensorflow as tf
from tensorflow.keras.models import load_model

folder = r'\\rdnas'
os.chdir(folder)

#os.environ["TF_KERAS"]='1'
model_loc = folder+'\\model.h5'
model = tf.keras.models.load_model(model_loc)
model.summary()  

# Onnx covnersion  

onnx_model = keras2onnx.convert_keras(model)
temp_model_file = 'model.onnx'
onnx.save_model(onnx_model, temp_model_file)

When I run the code I get the following error

Using TensorFlow backend.
Can't import tf2onnx module, so the conversion on a model with any custom/lambda layer will fail!
Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
DVA_input (Dense)            (None, 512)               670720    
_________________________________________________________________
dropout_3 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 256)               131328    
_________________________________________________________________
dropout_4 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_4 (Dense)              (None, 128)               32896     
_________________________________________________________________
dropout_5 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 10)                1290      
_________________________________________________________________
Predicted_Volume (Dense)     (None, 1)                 11        
=================================================================
Total params: 836,245
Trainable params: 836,245
Non-trainable params: 0
_________________________________________________________________
Traceback (most recent call last):

  File "<ipython-input-1-f9d072fc6a73>", line 19, in <module>
    onnx_model = keras2onnx.convert_keras(model)

  File "C:\Users\JTBLONIGAN\AppData\Local\Continuum\anaconda3\envs\Keras-GPU\lib\site-packages\keras2onnx\main.py", line 67, in convert_keras
    " Please set environment variable TF_KERAS = 1.")

Exception: This is a tensorflow keras model, but keras standalone converter is used. Please set environment variable TF_KERAS = 1.

I tried using the commented out line

os.environ["TF_KERAS"]='1' but nothing seemed to happen.

Has anyone had this issue or know how to change that variable to convert the model? Appreciate any help.

keras2onnx==1.6.0
Keras==2.2.4
onnx==1.6.0
tensorflow==2.0.0

回答1:

Environment variables should be set before importing modules, as modules check the variables at import:

import os
os.environ['TF_KERAS'] = '1'
import onnx

来源：https://stackoverflow.com/questions/58984892/how-to-set-environment-variable-tf-keras-1-for-onnx-conversion

标签

python

tensorflow

keras

spyder

ONNX