README

Keras-based YOLO v3, baseline solution for rebar-detection on Data-Foundation. It can also be used for VOC dataset.

Step 1. Install required packages.

pip install requirements.txt

Step 2. Generate VOC dataset. It would not be necessary if you already have one.

python toXML/generate_txt.py
python toXML/main.py

Step 3. Gererate augmented dataset with h5py file format. This step contributes to speeding up the training procedure.

python generate_h5_data_file.py

Step 4. Download YOLO v3 weights file from YOLO website, and convert it to h5 file format.

mkdir pretrined_weights
python convert2keras_weights.py yolov3.cfg yolov3.weights pretrined_weights/yolo.h5

Step 5. Train the model, and fine-tune if needed.

mkdir model_data_stage_1 model_data_stage_2
python stage_1.py
python stage_2.py

Step 6. Generate the predicted result.

python predict.py
python output.py

isee15/yolov3-keras