An autoencoder based approach to enable high fidelity video conferencing over low bandwidth networks

Abstract

Data lagging and distortion has been the issue for the majority of us since the usage of online video conferencing platforms has become a routine of our daily life. In this paper, we attempted to design a solution for this by specifically for the image part of the videos, by building a convolution neural network based autoencoder, which will compress the images being sent from one end to another in a batch of 5 frames, and stretch it back to its original size on the receiver end. We calculated the accuracy and loss obtained for the same for comparison purposes.