学习RCNN,SPPNet,Fast RCNN,Faster RCNN,R-FCN。这一系列是个递进关系,也是目标检测使用two-stage方法的一个发展过程。想要更好的理解Faster RCNN和R-FCN,只能把这算法都梳理清楚了,才能明白这个递进过程。
1.选择性搜索对象识别(Selective Search for object recognition)
首先要明白物体的检测框怎么界定,检测框的标定用了论文《Selective Search for object recognition》的方法,https://blog.csdn.net/sgfmby1994/article/details/78451748以及https://www.cnblogs.com/zhao441354231/p/5941190.html(这篇解读比较好)中有这篇论文的解读。
2.R-CNN
论文为《Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation》,论文网址为https://ieeexplore.ieee.org/abstract/document/6909475/metrics#metrics,论文的翻译链接https://blog.csdn.net/v1_vivian/article/details/78599229。论文提到的边框回归(BBR)问题参考博客https://blog.csdn.net/v1_vivian/article/details/80292569与博客https://blog.csdn.net/love1055259415/article/details/80041936
3.SPP-Net
论文为《Spatial Pyramid Pooling in Deep ConvolutionalNetworks for Visual Recognition》,论文网址为https://ieeexplore.ieee.org/abstract/document/7005506/authors#authors,论文解读https://blog.csdn.net/v1_vivian/article/details/73275259
其他目标检测的方法参考https://blog.csdn.net/linolzhang/article/details/54344350
4.Fast RCNN
论文为《Fast R-CNN》,论文网址为https://ieeexplore.ieee.org/abstract/document/7410526
5Faster RCNN
论文为《Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks》论文翻译网址为https://blog.csdn.net/quincuntial/article/details/79132243
原文:https://www.cnblogs.com/smy123/p/10480120.html