Feature/new onnx models

[2320sharon]

Hi Mark,
I created onnx models from the 3 tensorflow models you provided. Onnx is open-source format that acts like a "universal translator" for AI models, allowing them to be trained in one framework (like PyTorch or TensorFlow) and be used in a variety of software and hardware.

I validated that these new onnx models have the same outputs as the original tensorflow models. For example the total difference between the model score across an entire dataset was

max_abs_diff_overall: 0.0000076294
mean_abs_diff_overall: 0.0000033190

Using onnx will make your models accessible to more people since they won't have to have tensorflow installed to run them. They only need onnxruntime.

If you'd like I can submit a new version of the image_filter.py script that can be used to run the onnx models. I can call it something like image_filter_onnx.py, so both scripts are available to users.