Introduction

Scene parsing
Scene parsing is to segment and parse an image into different image region associated with semantic catgories.

• Most of scene parsing models use DNN to deal with pixel-wised classification problem. However, this method has problem in distingushing the categories with similar appereance.
• Therefore, this paper solved that problem from two aspects. 1: proprose an integreated classificiation model for scence parsing to distinguish confusing catgories/ 2: propose variance-based regularization to differentiate the scores of all categories as large as possible
• methos is in three steps
  1. Encoding features from DNN model
  2. general classification
  3. refining the classifier for refining the scores. -> for differentiating similar categories variance based regularization is used to train the intregrated classification model

=> Problem of previous segmentation model

References

• http://sceneparsing.csail.mit.edu/