Dear All,
I am currently working with a unique dataset that comes with 26 channel binary masks (26,256,256) for labels to segment arteries. Given that there exists an overlap between these classes, I am cautious about using the argmax approach to convert them into 2D masks. This is due to the potential risk of losing crucial information in regions where pixel values overlap.
I am reaching out to inquire if anyone has experience or insights on how to effectively handle such a scenario within the Fastai datablock for segmentation.
Thank you in advance for your time and valuable input.
Warm regards,
Bilal