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:

  1. R-CNN学习总结 https://zhuanlan.zhihu.com/p/30316608
  2. R-CNN论文详解 https://blog.csdn.net/WoPawn/article/details/52133338 (讲懂了 bounding box 回归)
  3. R-CNN 论文 http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=AF8817DD0F70B32AA08B2ECBBA8099FA?doi=10.1.1.715.2453&rep=rep1&type=pdf
  4. 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. 一次CNN计算完成以前 2k 次计算,, 然后用ROI Pooling把region proposal 缩放到固定大小 . 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:

  1. 【目标检测】Fast RCNN算法详解 https://blog.csdn.net/forever__1234/article/details/79919994
  2. Review: Fast R-CNN (Object Detection)

  3. https://www.mihaileric.com/posts/fast-object-detection-with-fast-rcnn/
  4. https://jhui.github.io/2017/03/15/Fast-R-CNN-and-Faster-R-CNN/
  5. http://www.robots.ox.ac.uk/~tvg/publications/talks/fast-rcnn-slides.pdf

 

 

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

 

 我对整个过程的理解:不用理解了,直接参考【5】就对了

 

 

 

 Ref:

  1. https://www.jianshu.com/p/ab1ebddf5
  2. 目标检测网络之Faster-rcnn解读(二)
  3. https://blog.csdn.net/hejin_some/article/details/88735023
  4. https://zhuanlan.zhihu.com/p/66228313
  5. 【程序喵笔记】目标识别1.0: Faster RCNN (这个写的很清楚,总算看懂了,也讲清楚了两次 NMS)

 

Mask R-CNN 2017

 

Mask R-CNN 提供两种backbone, with ResNet 和 with  FPN, 对应的的两种Head如下

 

 

 

 

图像分割算法

 Ref

  1. https://engineering.matterport.com/splash-of-color-instance-segmentation-with-mask-r-cnn-and-tensorflow-7c761e238b46
  2. https://blog.csdn.net/hejin_some/article/details/88735023

 

Overall:

 

 

 

Ref

  1.  Object Detection for Dummies Part 3: R-CNN Family
  2. https://www.itdaan.com/blog/2017/12/21/cf1f0f10ec7be98056bf126d5185ba57.html

 

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]

 

posted @ 2019-10-26 21:38  mashuai_191  阅读(407)  评论(0编辑  收藏  举报