Object detection

Autonomous Machines Jetson & Embedded Systems Jetson TK1
#1

Have anybody run successful a object detection on the Jeston TK1?
I try to run py-faster-rcnn but when i try to run it on the tk1 i get errors. On my PC it works fine.
I now the performance is not good on the tk1 but 1 FPS is enough for my work.

Thanks for help.
have a nice day and greetz from germany.
Torben

#2

Hi,

It’s not recommended to use TK1 for deep learning use case.
The GPU is too old and lots of libraries didn’t support it since 32-bit system.

Please buy a newer Jetson to unblock your issue.
Thanks.

#3

I’ve successfully ported darknet to tk1 which allows me to run YOLO3 on it. I can achieve 5fps using yolo3-tiny at 320x320 resolution which is sufficient for face detection.
My branch of darknet for tk1 is here: https://github.com/dtmoodie/darknet/tree/tk1

#4

Nice! Thanks for sharing!

#5

My branch of darknet for tk1 is here: https://github.com/dtmoodie/darknet/tree/tk1
I got an exception
make: *** No rule to make target obj/gemm.o', needed by libdarknet.so’. Stop.

#6

Can you check if there are any .o files in obj and verify that src/gemm.c exists? This is very strange.

#7

Thank you for your reply. I have successfully rebuild the project.
But when I have executed the command from tk1 branch:
./darknet detect cfg/yolov3.cfg /media/cluster_files/dev/weights/yolov3.weights data/dog.jpg
I have got only “asdf1” message. So the program has not found anything on the image

#8

Hello, I believe I’ve fixed the problem and you should be able to run object detection models trained on x86_64 systems correctly on the tk1.