优化策略（七）常用Head

Dense Prediction (one-stage):

(anchor based)

1. RPN
2. SSD
3. YOLO
4. RetinaNet

(anchor free)

1. CornerNet
2. CenterNet
3. MatrixNet
4. FCOS

Sparse Prediction (two-stage):

(anchor based)

1. Faster R-CNN
2. R-FCN
3. Mask RCNN

(anchor free)

1. RepPoints