matjesg / deepflash2

A deep-learning pipeline for segmentation of ambiguous microscopic images.
https://matjesg.github.io/deepflash2/
Apache License 2.0
58 stars 11 forks source link

Instance Segmentation Training #52

Closed pakiessling closed 1 year ago

pakiessling commented 1 year ago

Hi there, thank you for the tool. Excited to try it out.

I would like to use deepflash2 for instance segmentation. I have masks that look like this:

grafik

So far I have been using Cellpose. Do you think deepflash2 would be superior, or is it more focused on semantic segmentation? Do I have to convert my masks to binary foreground / background masks to train deepflash2?

matjesg commented 1 year ago

Hi @pakiessling deepflash2 can generally be used for similar use cases as cellpose -- it also uses cellpose in the background for instance segmentation, as described in our paper.

The instance segmentation masks are expected to have 0 as background and integers starting from 1 for each instance. Here are some examples how to convert the masks for some public datasets (conic, monuseg).