Why no different to run between yolov3 and yolov3-tiny on Jetson nano

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

Hi all

I deployed the yolov3 on Jetson nano follow those lines

sudo apt-get update
git clone GitHub - AlexeyAB/darknet: YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
cd darknet
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights
sudo vi Makefile
make
cd darknet

And no problem for running yolov3, but the issue is no different to run between yolov3 and yolov3-tiny, both around 1.7 fps.
./darknet detector demo cfg/coco.data cfg/yolov3.cfg yolov3.weights -c 0
./darknet detector demo cfg/coco.data cfg/yolov3-tiny.cfg yolov3-tiny.weights -c 0

yolov3-tiny should be around 15 fps, right?
Any idea how to fix it?

Your help is appreciated!

#3

Hi,

Please help to confirm that the app is really using the tiny model in the second run.
It sounds like it still use v3 standard.

Suppose there are some log information on the console.
Could you check it or share with us?

Thanks.

#4

Hi AastaLLL,

Could you tell me how can I find the log information on the console?

#5

1
12736×3648 3.48 MB

2
22736×3648 2.61 MB

Is that the information you need ?

#6

Could you find something wrong with the log information?

#7

Hi,

Since darknet use OpenCV for parsing video, it may not optimized for Jetson.
Would you mind to use our Deepstream SDK for an optimized multimedia pipeline?

We can get ~55fps with YOLOv3 Tiny model.

$ cd /opt/nvidia/deepstream/deepstream-5.0/sources/objectDetector_Yolo/
$ export CUDA_VER=10.2
$ make -C nvdsinfer_custom_impl_Yolo/

$ ./prebuild.sh 
$ export DISPLAY=:0
$ deepstream-app -c deepstream_app_config_yoloV3_tiny.txt

Thanks.