zaghlol94/yolo-v3

Yolo-V3

This is a pytorch implementation of Yolo V3, supported by pretrained weights for Pascal-VOC and inference script.

Setup And Inference

Code is developed under following library dependencies

python 3.7
torch 1.10.1
torchvision 0.11.2
opencv

Start with creating a virtual environment then open your terminal and follow the following steps:

git clone "https://github.com/zaghlol94/Yolo-V3"
cd Yolo-V3
pip install -r requirements.txt
bash download_assets.sh
cd src
python detect.py -I "image path"

Data-Set

This code we could use PASCAL_VOC or MS-COCO dataset

├── images
│   ├── 000001.jpg
│   ├── 000002.jpg
│   ├── 000003.jpg
│   ├── .
│   ├── .
│   ├── .
│   └── 009963.jpg
├── labels
│   ├── 000001.txt
│   ├── 000002.txt
│   ├── 000003.txt
│   ├── .
│   ├── .
│   ├── .
│   └── 009963.txt
├── test.csv
└── train.csv

For each image *.jpg in images folder there is a corresponding text file *.txt in the same name. Every line in the text file represents an object. Every line contains five values, the first value is the class id and the last four value is the bounding box. bounding box style is either midpoint style or corner

midpoint style example:-
11 0.34135977337110485 0.609 0.4164305949008499 0.262

Train on custom data

prepare your data set as the same structure of Data-set that showed in the previous section, furthermore change the DATASET and NUM_CLASSES variables and add list contains the classes name in config.py, then follow the following steps in the command line:

cd src
python train.py

Citation

@article{yolov3,
  title={YOLOv3: An Incremental Improvement},
  author={Redmon, Joseph and Farhadi, Ali},
  journal = {arXiv},
  year={2018}
}