The project attempts to classify 60000 images(50000 trainset and 10000 testset) from CIFAR-10 dataset into 10 classes: airplane ,automobile ,bird ,cat ,deer ,dog ,frog ,horse, ship or truck. This is done with the help of Convolutional Neural Netwroks(CNN).
The model structure was inspired from VGG-16. Several modifications were made to achieve a better accuracy.
The dataset can be previewed and downloaded from the link given below :
https://www.cs.toronto.edu/~kriz/cifar.html
Three Blocks have been employed in the classification task. Each block has a general form : CONV layers -> Relu -> Batch Normalisation followed by Max Pooling and Dropout. To enhance performance, Image Augmentation has been implemented. The model achieves 89.92 % Train accuracy and 89.78% Test accuracy.
The code is written in Python with the help of the popular deep learning framework Keras.