The twelve representative models we used are shown in the table。
Two Stage Models
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Mask R-CNN
Cascade R-CNN: High Quality Object Detection and Instance Segmentation
Cascade Mask R-CNN
RetinaNet
Hybrid Task Cascade for Instance Segmentation
One Stage Models
YOLOv3
SSD: Single Shot MultiBox Detector
YOLACT: Real-time Instance Segmentation
CenterNet
Transformer
DETR
Deformable DETR
