Semantic-Segmentation-using-Multiscale-L1-Loss

An Semantic Segmentation Network for segmenting medical WSI images using DeepLabv3 as Segmentor and Multiscale L1 Loss as Critic.

Papers:

• An integrated iterative annotation technique for easing neural network training in medical image analysis
• Rethinking Atrous Convolution for Semantic Image Segmentation
• SegAN: Adversarial Network with Multi-scale L1 Loss for Medical Image Segmentation

Label	Type	F1 Score	IoUs
0	Background	0.955	0.916
1	Glomureli	0.741	0.687
2	Podocyte	0.150	0.094
3	Non Podocyte	0.517	0.408