l4t-tensorRT image and cudnn

According to docs, of image :
https://ngc.nvidia.com/catalog/containers/nvidia:l4t-tensorrt
says in related docs:

For the latest TensorRT container Release Notes see the TensorRT Container Release Notes website.

I wonder… is that tensorrt the same tensorrt I am using nvidia:l4t-tensorrt. In that case, doesn’t exists release notes about that container. TensorRT docs says it container cudnn, but nvidia:l4t-tensorrt doesn’t. I mean, I cannot find it.

I only can found the so,


And that the package is installed:
root@agx1:/# apt list | grep libcudnn
libcudnn8/now 8.2.1.32-1+cuda10.2 arm64 [installed,local]
To this docker default installation, I can’t complete verification process:

conv_sample:

In file included from fp16_dev.h:15:0,
                 from fp16_dev.cu:13:
fp16_emu.h:61:10: fatal error: cudnn.h: No such file or directory
 #include <cudnn.h>
          ^~~~~~~~~
compilation terminated.
Makefile:203: recipe for target 'fp16_dev.o' failed
make: *** [fp16_dev.o] Error 1

Or with mnist:

/usr/bin/ld: cannot find -lcudart
/usr/bin/ld: cannot find -lcudnn
collect2: error: ld returned 1 exit status

And also, my problem is present:it’s not enough for compiling opencv. I would need these two requirements.

Because /cmake/modules/FindCuDNN.cmake of opencv doesn’t fint it and compilation says:

“# Could NOT find CUDNN (missing: CUDNN_LIBRARY CUDNN_INCLUDE_DIR)”

You need to install libcudnn8-dev on the system to get the cuDNN headers. I believe it will be mapped into the container as well if present on the system outside the container.

Ok, with your advice I could compile opencv. Now I need tensorRT(libinfer) dev, for compiling. tensorRT project.

But: