search

conscienceli / SeqNet

Joint Learning of Vessel Segmentation and Artery/Vein Classification
https://www.liangzhili.com/publication/li-2020-joint/
MIT License
61 stars 22 forks source link

result metrics #1

Open GinkgoX opened 4 years ago

GinkgoX commented 4 years ago

Dear author: The architecture is very powerful for both vessel segmentation and A/V classification, yet I have a question about the result metrics, could you explain why the segmentation use AUC/ACC metrics and how to get result as you have provided in parper: 'Joint Learning of Vessel Segmentation and Artery/VeinClassi cation with Post-processing'

Thanks a lot

conscienceli commented 4 years ago

It's a common way to use AUC/ACC as the metrics in vessel segmentation, as it is a binary classification problem. If you want the code for AUC/ACC, you can find it here: https://github.com/conscienceli/IterNet/blob/master/utils/evaluate.py You can easily merge them into SeqNet if you need them.

As for the metrics for A/V classification (full image/ all vessels/ major vessels/ etc.), I am using this one (https://github.com/rubenhx/av-segmentation/blob/master/evaluation_code.py). It's very convenient.