/DataAugmentationTF

Implementation of modern data augmentation techniques in TensorFlow 2.x to be used in your training pipeline.

Primary LanguageJupyter Notebook

Modern Data Augmentation Techniques for Computer Vision

About

Implementation of various data augmentation techniques in TensorFlow 2.x. They can be easily used in your training pipeline. This repository contain the supplementary notebooks for the Modern Data Augmentation Techniques for Computer Vision(Weights and Biases) report.

Techniques Covered

Note: Cutout, Mixup and CutMix are implememted in tf.data and can be found in the linked colab notebooks. I am using TensorFlow 2.x implementation of AugMix by Aakash Nain. His repo can be found here. The fork of this repo contains Weights and Biases integration and some additional command like arguments for more control.

Result

Check out the linked report for:

  • The comparative study of these augmentation techniques.
  • Augmentation implementations.
  • Evaluation of these augmentation techniques against Cifar-10-C dataset.

Model Used

ResNet-20