Quality of segmentation masks
#2
by
tcollyn
- opened
Hi there,
I was considering using your dataset for my masters thesis because it seemed extremely promising.
However upon inspecting the masks I found that some of them seem a bit dubious.
Masks like 57882471.jpg (and many others) seem very 'blocky' and do not seem to follow any natural edge of the wounds.
Sometimes the masks even seem to overlap on rulers used to measure the wounds in the image.
They also seem to not be binary masks (many pixel values present).
Could this be due to jpeg compression of the masks by any chance?
I would love to hear other opinions on the matter :))
(mask 57882471.jpg)
(The same mask that was binarized and overlayd over the corresponding image.)