Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Deepskin Model Error: Incorrect Inclusion of External Object in Wound Segmentation #5

Open
Skymero opened this issue Sep 25, 2024 · 2 comments

Comments

@Skymero
Copy link

Skymero commented Sep 25, 2024

An issue has been identified in the Deepskin model where it incorrectly includes a coin as part of the wound. As shown in the attached screenshot, the Deepskin model (outlined in green) has mistakenly incorporated a coin (circled in red) into its segmentation of the wound. This error highlights a potential flaw in the model's ability to accurately distinguish between the wound and external objects, which could have significant implications for its performance in real-world applications.

Screenshot 2024-09-25 111115

@Nico-Curti
Copy link
Owner

Dear @Skymero, thanks for your comments.
I totally agree with you that segmenting a red coin as a wound is a pitfall of the model.
However, we have already tested the model with different "coin markers" (in terms of color!) and the model works perfectly.
I imagine that the presence of a red coin near a human wound might be at least unusual for a real-world application...

If you are interested in placing circular markers near a wound as a landmark for further analysis, I strongly recommend using a reference with different colors: a good choice could be a green marker; we are currently using this kind of marker and hope to release a new version of the model that also supports marker segmentation in a few weeks.

@Skymero
Copy link
Author

Skymero commented Sep 26, 2024

I'd really love to help out with marker segmentation, is there any way I could help with that? I've already figure out how to measure the size of an object given a reference object. Currently it's not accurate since I just photoshopped a coin into an image I found and the FOV is assuming it was taken from a Galaxy s21 camera. So with more accurate data there I can adjust it.

Please let me know if I can help.

Here are other examples I ran into where it segmented parts of a ruler and the eyes on a face:

image

image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants