YOLO (You only look once) provides a framework for realtime object detection which works for images and videos. Because of the method how the image is detected it’s called YOLO.
YOLO is written based on Darknet (a open source neural network library).
More information about the method and how YOLO works:
https://pjreddie.com/darknet/yolo/
https://www.youtube.com/watch?v=4eIBisqx9_g
First test with YOLOv3
Download and prepare project darknet YOLO v3:
|
1 2 3 |
git clone https://github.com/pjreddie/darknet cd darknet make |
Download the weights a pre-trained model, which will be used to predict the objects.
|
1 |
wget https://pjreddie.com/media/files/yolov3.weights |
Default threshold for confidence is set to 0.25 but can be changed with parameter -thresh.
Test with a first custom picture:
|
1 |
./darknet detect cfg/yolov3.cfg yolov3.weights '/home/user/Desktop/cats.jpg' |
Test with other weights
|
1 2 |
wget https://pjreddie.com/media/files/yolov3-tiny.weights ./darknet detect cfg/yolov3-tiny.cfg yolov3-tiny.weights '/home/user/Desktop/cats.jpg' |