add: for tf 1.x & tf 2.x

[hirahara-taiki]

It can now be saved in SavedModel format for conversion to ONNX or TensorRT.

[gordanLiang]

It can now be saved in SavedModel format for conversion to ONNX or TensorRT.
can you teach me how to replace the current WHEnet(using tensorflow1.12.0)?

[shijh]

@hirahara-taiki excellent work. great appreciations!!!

I tested whenettf.py on python 3.8.10, tensorflow 2.3.4 on Windows 10 and works charmly. I used keras2onnx to save model to onnx format and load/test the onnx model with the 2 images under Sample folder, exact same results as h5's.

[gordinmitya]

thnx a lot!

with latest TF version (2.8.0) and python (3.9.7) I did:

from whenettf2 import WHENet
model = WHENet('WHENet.h5')
model.save('whened_sm')

and then

python -m tf2onnx.convert --opset 13 \
    --saved-model whenet_sm \
    --output whenet.onnx

otherwise I failed to convert original model even with old tf and python==3.6
zipped whenet.onnx