YOLO Augmentator is the project that augmentation methods for object detection dataset are implemented.
In this repositoty, I implemented 6 augmentation methods.
You can easily check and understand the results of each method through the visualization.
Sometimes the objects in image are severely cropped or occluded by augmentation. In that case, the augmented data may harm the stability of training and performance of the network.
To overcome this problem, the program check the IoU between non-augmented object and augmented object after augmentation, and then if there are objects that hardly loss their information, the program ignore that augmentation.
[Code Line 첨부]