Before solving the issue, I found another bug in Conv2DTranspose.build(). The len() cannot be used for input_shape after the commit 747e658, which change the type of input_shape from Shape to KerasShapesWrapper. I solved it by using len(single_shape) instead of len(input_shape), where single_shape = input_shape.ToSingleShape().
Back to the issue, the NotImplementedException raised because the lack of implement to handle 'use_bias' in 'Call' process of Conv2DTranspose. I add the code referring original TensorFlow.
Solve 2nd problem of #1012
Before solving the issue, I found another bug in Conv2DTranspose.build(). The len() cannot be used for input_shape after the commit 747e658, which change the type of input_shape from Shape to KerasShapesWrapper. I solved it by using len(single_shape) instead of len(input_shape), where single_shape = input_shape.ToSingleShape().
Back to the issue, the NotImplementedException raised because the lack of implement to handle 'use_bias' in 'Call' process of Conv2DTranspose. I add the code referring original TensorFlow.
Thanks!