Deep Reinforcement Learning in Computer Vision

DRL in CV Website

Google Slides on this introduction

In recent years, while use of Computer Vision techniques/models has burgeoned for solving Reinforcement Learning task(such as games), the opposite flow, of using techinques/models from Reinforcement Learning to solve paradigms in Computer Vision has also been seen.

Additionally, from a few stalwarts of Computer Vision:

This indicates that just as researchers in Reinforcement learning benifited from understanding and applying Computer vision techniques, researchers in Computer Vision can benifit from not treating Reinforcement learning as an esoteric black box and gaining a comprehensive understanding of this subject.

Hence, we are presenting a short series of lectures,(at our lab) with the following motivation:

DRL in CV Papers

An Additional repository has been made DRL_in_CV_Papers, which consist of a list of published works in computer vision which use Deep Reinforcement learning. A few of the papers have an added blog-post on them as well, highlighting important parts of the paper.

Posts

Additionally, for some topics which are important but might not have been a good slide presentation, we have made blog-like posts. This section will see further additions. It is open for additional posts from all. Kindly look in the _post folder for more information.

Acknowledgement

We rely heavily on the following for the content. This work is mostly curation of the excellant material already provided by these brilliant creators:

• Reinforcement Learning: An Introduction- Book by Andrew Barto and Richard S. Sutton. Link to latest draft.
• UCL course on RL - Course by David Silver.Link to material.
• Code material from various amazing sources: DannyBritz, ShangtongZhang, AndrejKarapathy.

This work has been complied by Aditya Ganeshan and Trisha Mittal while working at Video Analytics Lab(VAL),IISc. We thank the lab for giving us this opportunity.

Tutorials are still to be added for most chapters.