I am using Keras with TF backend for the training, the conversion from PB to UFF format is successful but the Uffparser throws an error at a concatenation layer with input (?, x0), (?, x1), (?, x2), (?, x3) and output (?, (x0+x1+x2+x3)) with ? being the input batch size. Concat axis is 1.
Exact parser error:
[TensorRT] ERROR: concatenate_1/concat: all concat input tensors must have the same dimensions except on the concatenation axis
[TensorRT] ERROR: UFFParser: Parser error: dense_2/BiasAdd: The input to the Scale Layer is required to have a minimum of 3 dimensions.
I guess the second error is related to the first one since the dense layer follows the concat layer.
Does TensorRT have problems with that variable size dimension on axis 0. If true, what could be a workaround?
TensorRT version 5, Python API.
I am happy to provide further information if needed.