Extract contour from segmentation output

[hannahorn]

Hello and thank you for your work,

I have a question regarding the post processing for the segmentation output.

In your paper you write: "we use a function to find contours of segmentation output".

What is done in cases where multiple contours are found for a given segmentation output?

Is this used as a flag to disregard the output as faulty, is the contour with greatest area always taken for further processing, or are all the contours combined in some manner?

[simongeek]

Hi,

Thanks for your interest in my work.

For HC (head circumference) and AC (abdominal circumference), we got two outputs: multi-class classification (head, abdomen, femur, or background), and segmentation maps (binary map). Then, if the binary map is like an ellipse, you are able to find the proper contour and make a rounded ellipse further. If the output has the wrong class (like femur, background) there is no possibility of having multiple contours, otherwise, we classify them as background.

I hope that it helps.

Please let me know, how my code can help you, and what kind of project are you working on.

Best wishes, Szymon