How to install Openpose on JetsonNano?

Since JetsonNano is a brand-new product,l am trying to install openpose on Nano generally based on the instruction for Jetson TX1 as follows:
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/installation_jetson_tx1.md
At last l can compile the whole project,but one error occurs,saying “cudaSuccess(48 vs.0 ) no kernal image is available for execuation on the device”.Maybe it is because we don’t have a webcam,but we gave a vedio’s address in TF card.
I supposed it is the problem of Nano’s calculation capability,but according to the configuration for Nano,it isn’t.Also,an official report says “Nvidia claims that the JetsonNano is caoable of running almost all modern AI,including Openpose and Tiny Yolo”
So ,what may cause this problem,and how to fix it?Thanks!

Hi,

Nano has the same compute capability with TX1. Both are 5.3 so it is not an issue.
A possible cause may be that the OS is updated to a newer kernel version.

So would you mind to update this command to kernel-4.9 and give it a try?

$ sudo cp /usr/src/kernel/<b>kernel-4.9</b>/arch/arm64/boot/Image ($PATH_TO_EMMC)/boot/Image

Thanks.

Hello,

I am trying to do the exact same thing and ran into the exact same problem.

The issue for me was indeed due the lack of the 5.3 compute capability in my caffe makefile.
I solved the issue by adding the following to my <CAFFE_DIR>/Makefile.config:

-gencode arch=compute_53,code=sm_53 \

I then ran (in <CAFFE_DIR>):

make clean
make all
make test
make runtest

Hope that helps

@nathanvw3:
what fps rate could you get?

The size of swapfile is 4G.

sudo swapon swapfile

run
./build/examples/openpose/openpose.bin -camera_resolution 640x480 -net_resolution 128x96

OK

but

./build/examples/openpose/openpose.bin --face -face_net_resolution 256x256

fail,

The Computer desktop is dead

@Andrey1984

I am still getting very low FPS. When I run the video example without any custom flags:

./build/examples/openpose/openpose.bin --video examples/media/video.avi

I get an FPS of around 0.4 :(

I’m currently trying to find ways of speeding this up without completely destroying accuracy. Currently trying to rebuild OpenCV.

If you have any suggestions I would love to hear them!