Please enable JavaScript.
Coggle requires JavaScript to display documents.
U-Net: Convolutional Networks for Biomedical Image Segmentation (Training,…
-
-
-
-
Methodology
Network Architecture 
modify and extend, few images, more precise segmentation
supplement a usual contracting network by successive layers where pooling operators are replaced by upsampling operators
high resolution features from the contracting path are combined (copy and crop)with the upsampled output
-
-
-
Training
-
Large Image, Small Batch=1
-
weight map for 1: different frequency 2: force to learn the separation borders between touching cells
-
Data Augmentation: shift and rotation invariance as well as robustness to deformation and gray value variations. displacement vectors on a coarse 3x3 grid. and sample from a gaussian distribution with 10 pixels standard deviation then bicubic interpolation, dropout layers at the end performa further implicit data augmentation