HyperNet, RON, FPN

HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection

主要是融合了不同层级的特征，比较容易理解，参考：
https://blog.****.net/u012905422/article/details/52614192
https://blog.****.net/u012361214/article/details/51374012

其中3.5. Speeding up的问题其实就是rfcn和Light head解决的问题，即把更多的计算量移动到roi pooling前面，以使得RoI之间尽可能的共享计算。

RON: Reverse Connection with Objectness Prior Networks for Object Detection
(a) multi-scale object localization
reverse connection, which enables the network to detect objects
on multi-levels of CNNs

个人感觉和FPN的路子很像。

(b) negative sample mining
propose the objectness prior to significantly reduce the searching
space of objects

Feature Pyramid Networks for Object Detection