DroneNet is Joseph Redmon's YOLO real-time object detection system retrained on 2664 images of DJI drones, labeled.
The original and labeled images used for retraining can be found under the image and label folders respectively.
-
Install the Ubuntu Linux distribution.
-
Open terminal and enter the following lines to build Darknet:
git clone https://github.com/pjreddie/darknet.git
cd darknet
make
Note: if you are using another variant of Darknet (e.g AlexeyAB fork), the labels are in a different format. Original Yolo uses absolute coordinates to indicate the target box in the image, whereas AlexeyAB (and others) adopts relative coordinates. Depending on your version, the appropriate labels can be found in the directory
labels
(absolute coords) ornormalized-labels
(relative). Of course, images remain the same. |
-
Move drone.data, drone.names, and yolo-drone.cfg under the cfg folder, create a weights directory and move yolo-drone.weights into the folder, move drone.jpg under the data folder, and move test.txt and train.txt in the root directory of your cloned darknet.
-
Change lines 2 and 3 to your path in drone.data.
Open terminal in the root directory of the darknet executable and enter:
./darknet detector test cfg/drone.data cfg/yolo-drone.cfg weights/yolo-drone.weights data/drone.jpg