fatal error: NvCaffeParser.h: No such file or directory~

[Nuwan1654]

I am getting the above error when trying to compile, any idea why?

[marcoslucianops]

Are you using docker?

[Nuwan1654]

yes, I pulled the docker using this command docker pull nvcr.io/nvidia/deepstream:6.2-samples

I am running it on jetson AGX Orin

[marcoslucianops]

Please run the command /opt/nvidia/deepstream/deepstream/user_additional_install.sh to install the dependencies before run the deepstream-app command.

[Nuwan1654]

I installed the dependencies with the above command, but the error is still there, also it is coming when I try to run CUDA_VER=11.4 make -C nvdsinfer_custom_impl_Yolo command

[Hasan-atc]

Hello, I was having the same problem and I made a make (in the Deepstream Yolo ONNX file) as described in this https://github.com/bharath5673/Deepstream and copied the DeepStream-Yolo file to my deepstream docker file. And when I run CUDA_VER=11.6 make -C nvdsinfer_custom_impl_Yolo again and run it, the problem is gone.

[marcoslucianops]

Please check with the updated files in this repo.

[pktiuk]

This problem may happen after installation of tensorrt in docker (Jetson Orin with Jetpack 6.0)

Reproduction Dockerfile

FROM nvcr.io/nvidia/deepstream:7.0-triton-multiarch

#REMOVING this line fixes compilation
RUN apt install -y tensorrt 

RUN cd /opt/nvidia/deepstream/deepstream/sources && \
git clone https://github.com/marcoslucianops/DeepStream-Yolo.git && \
cd DeepStream-Yolo/nvdsinfer_custom_impl_Yolo && \
export CUDA_VER=$(ls /usr/local | grep cuda- | tail -n 1| sed 's/cuda-//g') && \
echo "🟢 Found CUDA  version: $CUDA_VER (dir /usr/local/`ls /usr/local | grep cuda- | tail -n 1`)" && \
make

I created Issue on Nvidia Forum: https://forums.developer.nvidia.com/t/installation-of-tensorrt-package-removes-file-nvcaffeparser-h/295792

[marcoslucianops]

The TensorRT is already installed on this docker image.