I try to use TensorRT3 for inference of a Semantic Segmentation model, implemented in Tensorflow. After erasing all layers necessary for training (Dropout etc.), the UFF parer can convert my .pb model. When I then try to create the engine, it gives me an error on the output shape of a transposed convolutional layer.
[TensorRT] ERROR: UFFParser: Parser error: 2xfc8_1x1/z: Output shape of UFF ConvTranspose is wrong [TensorRT] ERROR: Failed to parse UFF model stream
Can you maybe help on this problem? Thanks a lot :)