Exported saved_model.pb in tensorflow 1.x with op from tf.contrib "ImageProjectiveTransformV2" doesn't work with tf-addons

[Bakaizhamgyrchiev]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):fedora 30
• TensorFlow version and how it was installed (source or binary):binary
• TensorFlow-Addons version and how it was installed (source or binary):binary
• Python version:3.7
• Is GPU used? (yes/no):no

I exported in tf 1.x saved_model. It includes op tf.contrib.image.transform ("ImageProjectiveTransformV2")

In tf 1.x before loading this saved_model I always import tensorflow.contrib (lazy loading)

So, I can't load this saved_model in tf 2.0 with importing tensorflow-addons
There is error:

NotFoundError: Op type not registered 'ImageProjectiveTransformV2' in binary running on train2.oz-services.ru. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) tf.contrib.resampler should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.

I have found that in

addons/tensorflow_addons/custom_ops/image/cc/kernels/image_projective_transform_op.cc

Line 162 in 22abf3c

REGISTER_KERNEL_BUILDER(Name("Addons>ImageProjectiveTransformV2") \

name "ImageProjectiveTransformV2" was replaced with "Addons>ImageProjectiveTransformV2".

Does it mean that I can't load this saved_model in tf2.0 and tensorflow-addons?

And if I add image custom ops from tensorflow-addons to tensorflow serving BUILD it also doesnt work with previous error

I watched #86

[Bakaizhamgyrchiev]

So, I understood that I should use tensorflow-addons<0.6

[seanpmorgan]

Thanks @Bakaizhamgyrchiev per this new RFC:
tensorflow/community#126

We've been asked to append package names on all of our custom ops to avoid collisions. Unfortunately this will be an issue for models with the older op names. as you mentioned TFA <0.6 should have compatible names.