multiple disease classes encoding and evaluation

[김선우]

in the case of multiple disease classes, when one pixel is masked BE and suspicious, which encoding form is correct?

[BE, suspicious, HGD, cancer, polyp ] = [1, 1, 0, 0, 0], is it right? 

and then, how does the evaluate this encoding form?

If the image has some pixels which have BE and suspicious, evaluate the prediction for both BE and suspicious ground truth? 

and what is the optimal value in the above case?

thanks.

[sali]

Hi,

We take VOC format bounding box input (in .txt format)

BE 0 25 45 345 456

suspicious 0 25 45 345 456

We take masks as multichannel for semantic segmentation where each channel refers to class

See wiki for details: 

 https://github.com/sharibox/EndoCV2020/wiki