metadata

library_name: keras
tags:
  - data-augmentation
  - image-classification

Tensorflow Keras implementation of : CutMix data augmentation for image classification

The full credit goes to: Sayan Nath

The Data augmentation strategy

CutMix is a data Augmentation strategy where some portion of the training example is removed and pasted with the content from other images in the training set. The labels are also mixed based on the proportion of the pixels that were combined. The full paper is at https://arxiv.org/pdf/1905.04899.pdf by Yun et. al., 2019.

CutMix augmented examples from CIFAR-10

Here are a few examples of augmented images.

Model and Dataset used

The Data augmentation is applied to the CIFAR-10 Data set. The model used here is the Resnet-20 with Categorical Cross-Entropy loss.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

optimizer: {'name': 'Adam', 'learning_rate': 0.001, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-07, 'amsgrad': False}
training_precision: float32

Training Metrics

Model history needed

Model Plot

View Model Plot