Download the best segmentation available for each image instead of random

[GalAvineri]

Some images have multiple segmentation masks available. As far as i've researched, their differ in their skill level.

Currently the system just chooses one of the masks, without consideration of the skill level. It would be preferred if there were an option to choose the highest skill level available

[neherh]

Maybe a better approach would be to be able to download all of the associated masks available and let the operator using the images to decide which masks are better. From what I have seen, sometimes the novice annotator may have a more defined segmentation mask while the expert creates an octagon around a mole and vice versa. Whether one approach is better than the other is clearly subjective based on what the person using the images determines to be correct.

[GalAvineri]

That is a sound idea :) When i'll get to implement this, i'll be sure to use this approach

[GalAvineri]

I've added the ability to mark a preferred segmentation skill level. :) 623b75c4

This means that when multiple segmentation are available, the script will download a segmentation with the preferred skill level. Otherwise, it will download a segmentation at random.

Plus, the names of the segmentation images now contain their skill level.

Can you confirm that you can this new functionality? :)

[MortenHannemose]

I second the idea of being able to download all segmentations associated with an image. :)

[GalAvineri]

This could be added by adding an additional option for the segmentation filter which is "no filtering".

On Wed, Oct 7, 2020, 20:56 Morten Hannemose [email protected] wrote:

I second the idea of being able to download all segmentations associated with an image. :)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/GalAvineri/ISIC-Archive-Downloader/issues/13#issuecomment-705099147, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFUF4HBBPIWKMBNKQI7QXDTSJSTWDANCNFSM4FLQZGLQ .