The codebase for:
- (Nov 2016): Zoom network with recursive training, arXiv link: [arXiv: 1702.05711]
- (July 2017): Zoom network with map attention decision, arXiv link: [arXiv: 1709.04347]
This codebase is written in Caffe with Matlab wrapper. It supports multi-gpu training. However, this is a quite old repo (active around early 2017) and I no longer maintain it.
Then why open-source it? Simply I have a habit to publish code whenever some work is accepted.