Notes for <Part-based R-CNNs for Fine-grained Category Detection>
Overview:
This paper is a nextention to R-CNN and it is mostly for Fine-grained Category Detection. It learns both object and part detectors and it also enforces learned geometric constraints between them.
- Region Proposals: selective search(same as R-CNN)
- find the bounding box for both the objects and parts, add the geometic constraints to them
