When I try to parse uff model in C++, I got the following error:
std::shared_ptr UffParser::parseConvTranspose(const uff::Node&, const Fields&, NodesMap&): Assertion `(outSpatial == layer->getOutput(0)->getDimensions().d && outSpatial == layer->getOutput(0)->getDimensions().d) && “Uff parser gets incorrect outSpacial”’ failed.
This error is caused by conv2d_transpose/conv2d_transpose op.
The uff model is converted from tensorflow pb file, and the original method from tensorflow is tf.layers.conv2d_transpose. I also list my usage of the function here:
upsampled = tf.layers.conv2d_transpose(inputs=input,
I tried to change kernel_size from 2 to 4 (not work).
I tried to change strides to 2 & 4 (not work).
Also, I found a similar Q&A by https://devtalk.nvidia.com/default/topic/1039218/tensorrt/uffparser-error-for-deconvolution-in-conv2d_transpose/
But the answer could not help me.
Linux distro and version: ubuntu 16.04
GPU type: P4000
nvidia driver version: 418.56
CUDA version 9.0
CUDNN version 7.1.4
Python version [if using python]
Tensorflow version 1.6
TensorRT version 5.1
What type of tf model/layers are you converting? tf.layers.conv2d_transpose