Question regarding the lesion segmentation ground truth quality

[alpapado]

First of all, thank you for releasing your dataset.

I was looking through the lesion segmentation images and couldn't help but notice that many ground truth images are completely black, even though the respective images contain obvious artifacts.

For example the below image clearly contains exudates but the corresponding exudate ground truth is an all black image.

This happens for many other images and lesion types as well. Please let me know what you think

[tim-murphy]

The patch of lesions to the left of the macula look like drusen and not exudate to me (see [1]). Drusen were not part of the dataset so wouldn't be included in the annotations. There are a couple of microaneurysms in the photo which were tagged correctly. Do you have a list of images where you think the annotations are missing?

I'm not part of this research project, just interested in it's data quality.

1: https://www.glycosmedia.com/education/diabetic-retinopathy/diabetic-retinopathy-normal-features-drusen/