Training AI to beat Pong game using Policy Gradients

I have made a 2 playered pong game in which both players are individual Neural networks and they are constantly trying to supress each other and maximize thier scores

Introduction

Both Neural Networks are given same image as an input, they give thier own ouputs respectively and action is taken on slider accordingly such that both try to maximize rewards. Both learn from thier experiences using policy gradients

Credits

The game was based on Siraj Raval's code and the structure of Neural Network was decided by taking reference to Andrej Karpathy's blog post