The CIFAR-10 Dataset consists of images of everyday objects. There are 50,000 training images (of which I've chosen 10,000 as validation images) and 10,000 test images, all of which are 32x32 pixels. More information about the dataset can be found at https://www.cs.toronto.edu/~kriz/cifar.html
I've used two different implementations. The first one (CIFAR-10_8052.ipynb) uses ConvNet along with Batch Normalisation and Callbacks (Reduce LR on Plateau) 80.52% accuracy is achieved. The second example (CIFAR-10_8392.ipynb) uses Image Data Generator on the above model. The accuracy is increased to 83.92%.
All implementations are made using Keras. The files are created in Google Colab. Link is included.
Comments and Suggestions are welcome :)
Author: Raahat Gupta Date: 02/09/2018