CIUS yolov4 training code

config step

See boyhoodme CSDN Blog

Large files downloads

• yolov4.conv.137
• yolov4.weights
• yolov4-tiny.conv.29 Download from Baiduyun, password: trdv These files are stored in SJUT_CIUS baiduyun

Running step

0. all default class is target, please keep the same
1. check your images in JPEGImages folder
2. check your xmls in Annotations folder
3. check ./data/train.data, ./data/train.names is set properly
4. check ./data/yolov4-tiny.cfg has changed classes and filters in 212, 220, 263, 269 lines
5. check the gen_files.py at line 10, the class name is right,such as target and then run

   python gen_files.py
   this python will clean the unlabelled images according to xml and generate txt files into ./labels
6. check the ./labels has the right txt file, open one to check.
7. copy the compiled darknet executable file from darknet projects, Note: we must use https://github.com/AlexeyAB/darknet.git master branch, Not https://github.com/pjreddie/darknet.git. Because AlexeyAB has the right version to train yolov4 or yolov4-tiny. But pjreddie only can train yolov3 or yolov3-tiny. 7.1 configure the makefile to only enable GPU=1 7.2 make clean && make 7.3 generate the darknet executable file
8. downloads large weight files from Baiduyun, password: trdv
9. see your tree is like

.
├── Annotations
├── backup
├── darknet
├── data
│   ├── train.data
│   ├── train.names
│   ├── yolov4.cfg
│   └── yolov4-tiny.cfg
├── gen_files.py
├── Images
├── JPEGImages
├── labels
├── README.md
├── test.txt
├── train
├── train.txt
├── yolov4.conv.137
├── yolov4-tiny.conv.29
└── yolov4.weights

10. run ./darknet detector train data/train.data data/yolov4-tiny.cfg yolov4-tiny.conv.29