CODEWITHSUNDEEP
pythontensorflowtensorflow-serving
Bakai Zhamgyrchiev
Bakai Zhamgyrchiev

Reputation: 23

Tensorflow serving with contrib operations

How can I serve model with tensorflow-serving, if there are tf.contrib operations. I use Tensorflow Serving via Docker (latest) (version of tf 1.11) and when I serve model there is the next message:

“Failed to start server. Error: Unknown: 1 servable(s) did not become available: {{{name: slider_universal version: 1} due to error: Not found: Op type not registered ‘ImageProjectiveTransformV2’ in binary running on 254345a5d9f1. 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 also built with bazel but there was the same error

I use tf.contrib.image.transform

If I delete this operation during exporting model it can be served by tensorflow serving

Upvotes: 2

Views: 1198

Answers (1)

John Pertoft
John Pertoft

Reputation: 1055

I had problems with the same op and it seems like the only way is to build it.

tensorflow_serving/model_servers/BUILD defines which Tensorflow ops will be included in variable SUPPORTED_TENSORFLOW_OPS, and I was a bit confused about this since it specifies that contrib ops should be included. However, since the tensorflow contrib build rule doesn't seem to include the ops under contrib.image so instead I explicitly added these by updating this variable to the following

SUPPORTED_TENSORFLOW_OPS = [
    "@org_tensorflow//tensorflow/contrib:contrib_kernels",
    "@org_tensorflow//tensorflow/contrib:contrib_ops_op_lib",
    "@org_tensorflow//tensorflow/contrib/image:image_ops_kernels",
    "@org_tensorflow//tensorflow/contrib/image:image_ops_op_lib",
]

Upvotes: 2

Related Questions