yiskw713 / paper_summary

Write summaries of papers I've read in Japanese
10 stars 1 forks source link

Learning Active Contour Models for Medical Image Segmentation #122

Open yiskw713 opened 4 years ago

yiskw713 commented 4 years ago

INFO

author

Xu Chen et al.

affiliation

conference or year

CVPR2019

link

pdf keras実装

概要

Active Contour Models の一つであるActive Contour Without Edges(ACWE)をセグメンテーションのロス関数として適用した研究.

提案手法

Active Contour Loss は,Length と Region の二つの項から成る. Length の項は,予測したセグメンテーションマップの total variation energy を表す. Region の項は,領域内/外の平均値c との二乗誤差の和をエネルギーとして表している. これらの和を最小化するように学習させる.

vがG.T.で,uが予測された確率マップを表す.

Screen Shot 2020-02-01 at 8 40 28

date

Feb. 1st, 2020