tensorflow test error as_list() is not defined on an unknow tensorshape

[Wayne]

 ERROR: test_multiple_batches_cpu (test_warpctc_op.WarpCTCTest)

---

Traceback (most recent call last):
File "/home/work/speech/libs/warp-ctc/tensorflow_binding/tests/test_warpctc_op.py", line 124, in test_m
ultiple_batches_cpu
self._test_multiple_batches(use_gpu=False)
File "/home/work/speech/libs/warp-ctc/tensorflow_binding/tests/test_warpctc_op.py", line 121, in test
multiple_batches
use_gpu=use_gpu)
File "/home/work/speech/libs/warp-ctc/tensorflow_binding/tests/test_warpctc_op.py", line 28, in _run_ct
c
self.assertShapeEqual(expected_costs, costs)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/test_util.py", line 1639, in a
ssertShapeEqual
np_array.shape, tf_tensor.get_shape().as_list(), msg=msg)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/tensor_shape.py", line 903, in
as_list
raise ValueError("as_list() is not defined on an unknown TensorShape.")
ValueError: as_list() is not defined on an unknown TensorShape.

and when I print cost's shape ,it's unknown shape.And if I eval cost,the shape and value is right.So What's the ctc return?
My env is tensorflow 1.10.1 and ubuntu 16.04

[Filip Ginter]

I had a similar problem with the missing shape and the latest tensorflow. The following modification to tensorflow_binding/src/warpctc_op.cc did the trick for me. You need to add the #include and then the .SetShapeFn... code and recompile. The code is copied and modified from here: https://www.tensorflow.org/guide/extend/op

#include "tensorflow/core/framework/shape_inference.h"

REGISTER_OP("WarpCTC")
    .Input("activations: float32")
    .Input("flat_labels: int32")
    .Input("label_lengths: int32")
    .Input("input_lengths: int32")
    .Attr("blank_label: int = 0")
    .Output("costs: float32")
    .Output("gradients: float32")
    .SetShapeFn([](::tensorflow::shape_inference::InferenceContext* c) {
       c->set_output(0, c->input(3));
       c->set_output(1, c->input(0));
       return ::tensorflow::Status::OK();
     });

[Nicolas Camenisch]

To get even better results I modified the answer by @fginter based on the original Tensorflow .SetShapeFn implementation of the CTCLoss function.

#include "tensorflow/core/framework/shape_inference.h"

using ::tensorflow::shape_inference::DimensionHandle;
using ::tensorflow::shape_inference::InferenceContext;
using ::tensorflow::shape_inference::ShapeHandle;
using ::tensorflow::Status;

REGISTER_OP("WarpCTC")
    .Input("activations: float32")
    .Input("flat_labels: int32")
    .Input("label_lengths: int32")
    .Input("input_lengths: int32")
    .Attr("blank_label: int = 0")
    .Output("costs: float32")
    .Output("gradients: float32")
    .SetShapeFn([](InferenceContext* c) {
        
        ShapeHandle activations;
        ShapeHandle flat_labels;
        ShapeHandle label_lengths;
        ShapeHandle input_lengths;
        
        TF_RETURN_IF_ERROR(c->WithRank(c->input(0), 3, &activations));
        TF_RETURN_IF_ERROR(c->WithRank(c->input(1), 1, &flat_labels));
        TF_RETURN_IF_ERROR(c->WithRank(c->input(2), 1, &label_lengths));
        TF_RETURN_IF_ERROR(c->WithRank(c->input(3), 1, &input_lengths));
        
        
        // Get batch size from inputs and sequence_length, and update inputs
        // with the merged batch_size since it is returned.
        DimensionHandle batch_size;
        TF_RETURN_IF_ERROR(
                           c->Merge(c->Dim(activations, 1), c->Dim(input_lengths, 0), &batch_size));
        TF_RETURN_IF_ERROR(c->ReplaceDim(activations, 1, batch_size, &activations));
        
        c->set_output(0, c->Vector(batch_size));
        c->set_output(1, activations);
        return Status::OK();
    });