目标检测框架差异

1）R(Regions)-CNN https://arxiv.org/abs/1311.2524
Girshick R, Donahue J, Darrell T, et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[C]// Computer Vision and Pattern Recognition. IEEE, 2013:580-587.

2）Fast R-CNN https://arxiv.org/abs/1504.08083
Girshick R. Fast R-CNN[C]// IEEE International Conference on Computer Vision. IEEE, 2015:1440-1448.

3）Faster R-CNN https://arxiv.org/abs/1506.01497
Girshick R. Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015: 1440-1448.

4）YOLO https://arxiv.org/abs/1506.02640
Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 779-788.

5）SSD https://arxiv.org/abs/1512.02325
Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision. Springer International Publishing, 2016: 21-37.

6）YOLOv2 https://arxiv.org/abs/1612.08242
Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[J]. arXiv preprint arXiv:1612.08242, 2016.

7）Mask R-CNN https://arxiv.org/abs/1703.06870
He K, Gkioxari G, Dollár P, et al. Mask R-CNN[J]. arXiv preprint arXiv:1703.06870, 2017.

R-CNN: region proposal可能是object的区域放到CNN获得分类；问题：区域重叠多，重复计算量大
Fast R-CNN: 多个可能的区域一起放到CNN，只做一次feature提取，获得分类；问题：获得region的速度太慢
Faster R-CNN: region proposal放到CNN里来完成，CNNCNN；问题：获得位置分类，两步走速度慢
YOLO：位置+分类回归问题，一步解决；问题：小物体、挨得很近的物体、尺寸不太常见的物体的漏检
SSD：借鉴Faster R-CNN和YOLO，增加几个对不同尺寸的感知层
YOLO2：调整特征提取部分的网络结构更快，使用归一化、针对细颗粒度、高分辨率进行优化，修改训练策略更准