yolo-v3-keras

Keras implementation of YOLO v3 https://pjreddie.com/media/files/papers/YOLOv3.pdf.

Pretrained model

You can download offitial pascal pretrained model here

Ussage

To make prediction use this command:

git clone https://github.com/gosha20777/yolo-v3-keras.git
cd yolo-v3-keras
wget -O voc.h5 https://github.com/gosha20777/yolo-v3-keras/releases/download/0.1.0/voc.h5
python predict.py -c config.json -i /path/to/image_dir -o /patch/to/output_dir

Training

1. Load the VOC training set http://host.robots.ox.ac.uk/pascal/VOC/voc2012/.

2. Edit config.json file and chnge train_image_folder, train_annot_folder, valid_image_folder and val_annot_folder patches.

3. Run: python train.py -c config.json

Training on your own data

1. Create your VOC-style dataset. (You can do it with labelImg program).
2. Edit classes at config.json.
3. Train it! I highly recommend you to use default pretrained weights and fine-tune it