NOTE: If you want to find the best data augmentation policy for your dataset, I suggest looking at https://paperswithcode.com/task/data-augmentation. This repository was mainly used as practice for learning software enginnering skills; it is not the most practically useful tool.
Contains
- A simple PyTorch compatible library implementing several algorithms for automatically finding good data augmentations for image datasets.
- An accompanying web app made with Flask and React
Documentation: https://autoaug.readthedocs.io/en/latest/
Report: https://github.com/sunjin000/auto-augmentation/blob/master/docs/source/explanation/report.pdf
See https://paperswithcode.com/sota/data-augmentation-on-imagenet for further elaboration (and the latest benchmarks) on this problem setting.
Made for the Software Engineering Project in the MSc Artificial Intelligence course at Imperial College London by John Carter, Maxim Ramsay King, Mia Yixuan Wang, Sunjin Kim. It's pretty simple and rudimentary, but we learned a lot about software engineering from making this.