Awesome

YOLOv3

Keras(TF backend) implementation of yolo v3 objects detection.

According to the paper YOLOv3: An Incremental Improvement.

Requirement

• OpenCV 3.4
• Python 3.6
• Tensorflow-gpu 1.5.0
• Keras 2.1.3

Quick start

• Download official yolov3.weights and put it on top floder of project.

• Run the follow command to convert darknet weight file to keras h5 file. The yad2k.py was modified from allanzelener/YAD2K.

python yad2k.py cfg\yolo.cfg yolov3.weights data\yolo.h5

• run follow command to show the demo. The result can be found in images\res\ floder.

python demo.py

Demo result

It can be seen that yolo v3 has a better classification ability than yolo v2.

<img width="400" height="350" src="/images/res/dog.jpg"/><img width="400" height="350" src="/images/res/person.jpg"/>

TODO

• Train the model.

Reference

@article{YOLOv3,  
  title={YOLOv3: An Incremental Improvement},  
  author={J Redmon, A Farhadi },
  year={2018}

Copyright

See LICENSE for details.