Tractable Large-Scale Deep Reinforcement Learning

Some results from the latest paper Tractable Large-Scale Deep Reinforcement Learning by Nima Sarang and Charalambos Poullis, on deep reinforcement learning. The code is publicly available at https://github.com/nsarang/road-extraction-rl.

TL;DR: We reformulate the problem of road extraction by training an RL agent to traverse the roads from satellite images. We introduce novel techniques for making this inherently complex problem tractable on commodity GPUs and propose a self-supervised loss and actions tailored for road extraction.

The orange lines show the ground truth and are only used during training. The first three videos are from the test dataset and the final from the training.

0:00
/
Montreal, QC, Canada (fast-forward, test)
0:00
/
Montreal, QC, Canada (test)
0:00
/
Los Angeles, CA, USA (fast-forward, test)
0:00
/
London, UK (train)

Subscribe to Charalambos Poullis

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
name@example.com
Subscribe