Faster R-CNN study notes

data record

Paper translation http://blog.csdn.net/liumaolincycle/article/details/48804687
faster-rcnn principle and corresponding concept explanation
http://www.cnblogs.com/dudumiaomiao/p/6560841.html

Problem Description

Object Recognition and Location: Determine whether an object exists in a picture (object recognition), and give its exact location (location).
Object localization generally gives a rectangle and the type of the object. The coordinates of the upper left corner of the rectangle representing the position of the object$(x_i,y_i)$, the width and height of the rectangle$(w_i,h_i)$, the type of object$c_i$, which forms an array. There may be multiple objects in a picture, so the recognition result is a set of five-tuples.

which is$y_i$is a quintuple.

$$y_i = \{(x_i,y_i,w_i,h_i,c_i)\}$$

notes

R-CNN first enumerates all candidate boxes in a picture, extracts each candidate box as a chapter picture, and puts it into the CNN network for learning.

EVERYTHING

To be continued