Patchmentation is a python library to perform patch augmentation, a data augmentation technique for object detection, that allows for the synthesis of new images by combining objects from one or more source images into a background image.
The process of patch augmentation involves extracting objects of interest from the source images, transforming them, and then pasting them onto the background image to create a composite image, therefore increasing diversity at the object level. The resulting dataset offers a greater variety of object combinations within a single image, making it more robust and accurate when training object detection models.
The easiest way to install patchmentation is through pip.
pip install patchmentation
Note: Some functionality of patchmentation might not be working on non-Linux systems.
-
GitHub Repository: https://github.com/Xu-Justin/patchmentation
-
Docs: TBA
-
Research Paper: TBA
-
Benchmarking Results: https://github.com/Xu-Justin/patchmentation-yolov5
This project was developed as part of thesis project, Computer Science, BINUS University.