We are trying to build a artificial agent that learns to count the number of objects in a scene. Previous, we built a simple agent bsaed on Q learning. Here, we are using a more sophisticated architecture: the recurrent model of attention [1], or RAM.

Our actual implementation of the RAM is here. In this repo, I am using this RAM model to do some counting related tasks.

References:

[1] https://arxiv.org/abs/1406.6247