By, Bachotti Sai Krishna Shanmukh, Undergraduate in Department of Electrical Engineering, IIT Madras
Professor : Nitin Chandrachoodan
Medical Images are used in healthcare to diagnose certain health conditions and often require a very accurate pixel map for computer-aided diagnosis. These images are also used as datasets for both supervised (or) unsupervised deep learning models to diagnose critical health conditions, analyze the nature and spread of tumors etc. Hence, compressing these images in a lossless way i.e., capturing the complete information present in the original pixel map into a lower file size configuration is critical for accurate diagnosis. In this paper, we discuss about using Neural Networks for lossless image compression.
This repository contains the code of four autoencoder models, weights and results.
project_report.pdf contains a detailed report of the problems addressed and the implementation
NN : Folder for ANN based Autoencoder trained on MSE loss
CNN : Folder for CNN based Autoencoder trained on MS_SSIM based loss
CNN2 : Folder for CNN based Autoencoder with increased latent space dim and trained on custom loss
CNN3: Dual Stage Autoencoder
HiFiC : Pre-Trained GAN model compression bit-rate
Dataset : Kaggle Chest X-ray Dataset