R-CNN, Fast R-CNN, Faster R-CNN, Mask R-CNN

最近在看 Mask R-CNN，这个分割算法是基于 Faster R-CNN 的，决定看一下这个 R-CNN 系列论文，好好理一下

R-CNN 2014

1. 论文 Rich feature hierarchies for accurate object detection and semantic segmentation Tech report (v5)

Author: Ross Girshick Jeff Donahue Trevor Darrell Jitendra Malik, UC Berkeley

link: https://arxiv.org/pdf/1311.2524.pdf (2014)

2. 【目标检测】R-CNN论文详解（Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation）https://www.jianshu.com/p/c1696c27abf8

Ref:

R-CNN学习总结 https://zhuanlan.zhihu.com/p/30316608
R-CNN论文详解 https://blog.csdn.net/WoPawn/article/details/52133338 (讲懂了 bounding box 回归)
R-CNN 论文 http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=AF8817DD0F70B32AA08B2ECBBA8099FA?doi=10.1.1.715.2453&rep=rep1&type=pdf
Review: R-CNN (Object Detection) https://medium.com/coinmonks/review-r-cnn-object-detection-b476aba290d1

Fast R-CNN 2015

R-CNN 有两个问题：1. Selective Search 发现的~2k个region proposal 都要被送到CNN feature extractor里，实际上这2k个region 有大量重复的部分，所以造成很多重复计算. 2. 有3个model 独立训练(feature extractor, SVM classifier, bbox refine)，占用大量时间

Solution： 1. 用ROI Pooling 一次计算完成以前 2k 次计算. 2. 合并以前的3个网络成一个整体. 用softmax代替了svm 分类器.

NOTE: 还是用到了 Select Search.

Ref:

Faster R-CNN 2016

Fast R-CNN 还是使用 Selective Search 先选取 region proposal, Faster R-CNN 使用叫RPN 的CNN网络选 region proposal

Mask R-CNN 2017

图像分割算法

Ref

https://engineering.matterport.com/splash-of-color-instance-segmentation-with-mask-r-cnn-and-tensorflow-7c761e238b46

Overall:

Ref

Object Detection for Dummies Part 3: R-CNN Family

A Brief History of CNNs in Image Segmentation: From R-CNN to Mask R-CNN

Here is a list of papers covered in this post ;)

Model	Goal	Resources
R-CNN	Object recognition	[paper][code]
Fast R-CNN	Object recognition	[paper][code]
Faster R-CNN	Object recognition	[paper][code]
Mask R-CNN	Image segmentation	[paper][code]

巴特西