Closed MichaelX99 closed 6 years ago
Wow, Learning models for visual 3D localization with implicit mapping does a better job explaining GQN than Neural scene representation and rendering did... All the hyperparams and architecture choices in the code make sense now
The DRAW module is cited and mentioned in the GQN paper, just under the name ConvDRAW Gregor et al 2016.
First off, thank you for posting this code as it is very helpful! Its not a problem but I am curious why you decided to use the DRAW module in your code even though it is nowhere cited in the GQN paper? Granted the paper lacks many implementation details...