While most steps in the modern object detection methods are learnable, the region feature extraction step remains largely handcrafted, featured by Rol pooling methods. This work proposes a general viewpoint that unifies existing region feature extraction methods and a novel method that is end-to-end learnable. The proposed method removes most heuristic choices and outperforms its Rol pooling counterparts. It moves further towards fully learnable object detection.
展开▼
