S.M.Ali Eslam et.al. Attend, Infer, Repeat: Fast Scene Understanding with Generative Models ICML16
This is a basic bare-bone implementation which does not include structured data-driven neural baselines that was used in the original paper to reduce variance in gradient estimates. Can be used for educational purposes and understand the paper better. TODO : Add more documentation PLease check out the Pyro tutorial on AIR for a very detailed explanation of the model.