YOLO-Pytorch

中文说明

Description

This is a pytorch version of YAD2K。

Original paper: YOLO9000: Better, Faster, Strongerby Joseph Redmond and Ali Farhadi.

Requirements

• Pytorch 0.3.0
• torchvision
• opencv（Requirement for camera and video）
• python 3

Usage

1. Download Darknet model cfg and weights from the official YOLO website.

   # for example --- or other version cfg and weights
   wget http://pjreddie.com/media/files/yolo.weights
   wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolo.cfg
   

   Note: you can download other types: like yolo-voc.cfg

2. Convert the weights to .pth

   python tools/yad2t.py path-to-yolo-cfg path-to-yolo-weights path-to-output-folder
   

   Note: default choose

   • copy your yolo.cfg and yolo.weights to the directory config
   • the output folder is model

3. Three demos (picture, camera, video)

   1. demo.py

      python demo.py pic-path yolo-type --cuda=True
      

      Note: default choose

      • picture in folder results/demo
      • yolo-type is yolo： three kinds: [yolo, tiny-yolo-voc, yolo-voc]

   2. demo_cam.py

      python demo_cam.py --trained_model=pth_model_from_1
      

   3. demo_video.py

      python demo_video.py --demo_path=video_path --trained_model=pth_model_from_1