Pick up the interpolation attributes of Keras's UpSampling2D layers.

[kohei-us]

In ONNX, An Upsample operator may have the mode value of either 'nearest'
or 'linear'. In Keras, UpSampling2D may have the interpolation value of
either 'nearest' or 'bilinear'.

[CLAassistant]

All committers have signed the CLA.

[jiafatom]

Thanks for the change. (1) Please add a unit test in test_layers.py for linear mode. (2) Please fix build failure for python 3.5.

[kohei-us]

Hmm... Obviously the version of Keras used in the linux-conda-ci agent does not support the interpolation argument in its UpSampling2D layer implementation...

[jiafatom]

Hmm... Obviously the version of Keras used in the linux-conda-ci agent does not support the interpolation argument in its UpSampling2D layer implementation...

Let's find since which keras version interpolation argument is supported. Then in unit test, we use
if StrictVersion(keras.version) >= StrictVersion()
to only enable this test for the keras version when interpolation argument is supported.

[jiafatom]

Just checked https://github.com/keras-team/keras/releases, Add interpolation argument in UpSampling2D layer , in keras 2.2.3

[kohei-us]

Thanks for your suggestion. I've made changes to check the version of Keras to conditionally disable my test case. Hopefully this will work.

[jiafatom]

sign CLA please?

[kohei-us]

sign CLA please?

I need to check with someone in our organization before signing this. Let me work on this, and I'll get back to you.

[jiafatom]

Also please fix the inline comments about write statements concisely.

[kohei-us]

Also please fix the inline comments about write statements concisely.

Done.

[jiafatom]

Since cla is stil pending, we apply the fix here instead. Thank you for your contribution.

[kohei-us]

I know it's too late, but I've finally got approval internally to sign the CLA.