Deep learning Experiments

I will cover few deep learning topics here and I will share my observations on each of these topics. The topics I intend to cover is specified here though it is not complete.

Topics covered

Classification :

A basic binary classifier for cat-dog classification using a Convolutional Classifier is learned here. Then we experiment with the label smoothing concept introduced for model calibration and improved accuracy.

Autoencoders :

A basic Multi-layer perceptron (MLP) based autoencoder is designed for MNIST dataset. We then check the effect of bottleneck layer on the reconstruction. A visualization of the data in pixel space as well as in encoded latent space with TSNE is provided. Then we move on to playing around with linear interpolation in the latent space and image space.

Types of Autoencoders (coming soon) :

I intend to cover contrastive AE and Variational AE to give a first step ino generative networks

GANs (coming soon) :

• DCGAN for face generation (with Latent space vector arithmetics)
• Pix2Pix/cycleGAN
• Impact of GAN in restoration framework (A simple SR/ denoising/ image inpainting )
• My own deblurring paper (http://openaccess.thecvf.com/content_ECCV_2018/papers/Nimisha_T_M_Unsupervised_Class-Specific_Deblurring_ECCV_2018_paper.pdf)

Hope to cover few of these ASAP