How to import Neural Networks from ONNX?

Question

ONNX is a open format to represent deep learning models.

So I want to import neural networks from other frameworks via ONNX.

And the Mathematica 11.3 supports python now.

I think I can use ONNX-MXNet to export the mxnet.json and mxnet.params then just Import these.

But I have no idea about how to install packages on Python-ExternalSessions.

---

Example

Here is an onnx-format file of Super Resolution Model.

How can I import this using Mathematica and Python-ExternalSessions?

Answer 1

This is incomplete answer because Mathematica doesn't support the latest version of mxnet.

Step 1

Install onnx-mxnet. Open Command Prompt and enter:

pip install onnx-mxnet

Step 2

Download model and image. Run this code in Mathematica:

import mxnet as mx
import onnx_mxnet
import numpy as np

sym, params = onnx_mxnet.import_model('C:/Users/Alexey/Documents/super_resolution.onnx')

from PIL import Image
img = Image.open('C:/Users/Alexey/Documents/super_res_input.jpg').resize((224, 224))
img_ycbcr = img.convert("YCbCr")
img_y, img_cb, img_cr = img_ycbcr.split()
test_image = np.array(img_y)[np.newaxis, np.newaxis, :, :]

mod = mx.mod.Module(symbol=sym, data_names=['input_0'], context=mx.cpu(), label_names=None)
mod.bind(for_training=False, data_shapes=[('input_0',test_image.shape)], label_shapes=None)
mod.set_params(arg_params=params, aux_params=params, allow_missing=True, allow_extra=True)

mod.save_checkpoint("super_resolution",1)


net = Import["super_resolution-symbol.json", "MXNet"]