network-for-cifar-10

This demonstrates to achieve 90% accuracy on CIFAR-10 dataset

Environment

• Tensorflow-gpu=2.0.0
• Keras=2.3.1
• 1 GPU with CUDA Version: 10.1

Network

• Convolution layer is a set of 3 operations: Convolution, Activation & Batch normalization. Contain 8 convolution layers, 4 max Pooling layers.
• The dropout layer is kept after the Pooling layer. I add 4 dropout layers with the rate is: 0.2, 0.3, 0.4, 0.5. Dropout used to prevent a model from overfitting.

Result:

• Result: 89.050% in test set
• Here is the original code: https://github.com/abhijeet3922/Object-recognition-CIFAR-10