Custom object detection with Yolov3 (paper), using AlexeyAB's framework. Trained on 800 images of Acer rubrum for 3500 batches, starting with the pre-trained yolov3.weights. Training set images downloaded from iNaturalist and labeled with LabelImg.
$ git clone https://github.com/etowahs/darknet.git
$ cd darknet
$ make <-- for Linux only
If you have CUDA installed on your computer, you may want to have GPU=1 in Makefile to have faster detection. See addtional instructions for Linux and Windows.
Download trained weights from here (235MB) and place it in the darknet folder.
To run the detector on a single image:
$ cd darknet
$ ./darknet detector test custom/darknet.data custom/yolov3.cfg leaf.weights -ext_output my-image.jpg
To run the detector on an entire folder of images, create a .txt file containing the file locations of all the images. Ex. my-imgs.txt
$ ./darknet detector test custom/darknet.data custom/yolov3.cfg leaf.weights -ext_output < my-imgs.txt > output.txt