This website requires JavaScript.

Persistent Homology with Improved Locality Information for more Effective Delineation

Doruk OnerAd\\'elie GarinMateusz Kozi\\'nskiKathryn HessPascal Fua
摘要
We present a new, more effective way to use Persistent Homology (PH), amethod to compare the topology of two data sets, for training deep networks todelineate road networks in aerial images and neuronal processes in microscopyscans. Its essence is in a novel filtration function, derived from a fusion oftwo existing techniques: thresholding-based filtration, previously used totrain deep networks to segment medical images, and filtration with heightfunctions, used before for comparison of 2D and 3D shapes. We experimentallydemonstrate that deep networks trained with our Persistent-Homology-based lossyield reconstructions of road networks and neuronal processes that preserve theconnectivity of the originals better than existing topological andnon-topological loss functions.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答