- Convert images to random affine, rotation, translation, stretching, color noise, brightness, etc.
- The default is only for target detection data enhancement, generating images and corresponding yolo format annotation files (image classification needs to modify the code part)
- You can specify the number N of generated data, but the actual number of generated data is about 0.8*N
- pip install opencv-python -i https://mirror.baidu.com/pypi/simple
- Based on the original image (data/imgs) and classification labels (data/labels.txt), use the online annotation tool https://www.makesense.ai/ to make a vgg format annotation file (data/vgg.json The file here already exists , the user can also recreate)
- Generate data The image data generated by python generator.py is stored in (fake/images), and the corresponding labels are stored in (fake/labels)
- Verify the generated data python plot.py draw the box of the annotation file to the corresponding generated image, and save the drawn image in (fake/images)
- If you need to convert yolo format to coco format, please refer to https://github.com/yfq512/DL_tools