Learning To Paint with RL

This problem aims at training an agent to generate paintings using stroke based sequential actions. We plan to break down the given target image into a set number of strokes, using a RL agent, which can generate a painted image. Compare On Policy (PPO) and Off Policy (DDPG) methods for solving the problem. We also aim at incorporating style transfer while painting.

Quickstart

For training and testing on MNIST

For training and testing on celeb

For training Style transfer RL on MNIST

Style transfer

Demo

Style Transfer on MNIST

Using DDPG on CelebA

Using PPO on MNIST

Version Info

Version : 1.0.0

Authors

• Asmit Kumar Singh - IIIT-Delhi - Other Work
• Suchet Aggarwal - IIIT-Delhi - Other Work
• Pragya Sethi - IIIT-Delhi - Other Work

Acknowledgments

• Code for DDPG and environment