Convolutional neural networks for image classification using the Torch library
We construct two Convolutional Neural Networks (CNN) for image classification, using the PyTorch library.
The data used comes from the widely popular CIFAR-10 dataset which consists of 60000 32x32 color images in 10 classes, with 6000 images per class.
The architecture of the networks are as follows:
The first CNN consists of:
- Two convolutional layers of size 3 with padding.
- ReLU activation and MaxPooling after each convolutional layer.
- Finally, a fully-connected Linear layer.
The second CNN consists of:
- Three convolutional layers of size 3 with padding.
- ReLU activation, MaxPooling and batch normalization after each convolutional layer.
- Three fully-connected Linear layers with ReLU activation.