Closed JadBatmobile closed 5 years ago
I tried TensorRT 5 first actually, and then i thought the function signature may have changed for 5, so i went to TensorRT4 and the same occurred
is the the signature it is supposed to have: virtual bool registerInput(const char* inputName, nvinfer1::Dims inputDims, UffInputOrder inputOrder)?
should i true tensorRT 3?
Hi,
Have you tried the branch jetpack4.1-jetson-xavier
?
I haven't had time to merge the branches and compile different api calls depending on the version of tensorRT. Actually it's pretty depressing that NVIDIA would make such a change, it seems unnecessary. I will add an issue so I remember to fix this.
This is no longer a problem for me. It wasn't the batch you suggested, but there were confilicting installations of tensorRT on my computer, i fixed it by wiping and reinstalling. This should be closed.
I see the same issue with TensorRT5
, I switched to the suggested branch, and solves the issue. Thanks
Hello,
I am attempting to use deploy_cpp with TensorRT backend.
I have installted TensorRT 4 succesfully,
when i run catkin_make, it throws the warning that i do not have tensorflow_cc, but that tensorRT is found. I am under the impression that should be enough?
The following is what i am seeing /home/ubuntu/bonnet_ws/src/bonnet/deploy_cpp/src/lib/src/netTRT.cpp:113:56: error: no matching function for call to ‘nvuffparser::IUffParser::registerInput(const char*, nvinfer1::DimsCHW&)’ _parser->registerInput(_input_node.c_str(), inputDims);