Question about the 2018 challenge

This question is about the 2018 challenge, but that forum is not active anymore. Sorry for spamming.

  1. Are lesions from Task1 and Task2 of the same classes as in HAM10000?
  2. Can we get ground-truth class labels for the lesions in Task 1 and Task 2 (similar to the classes in Task 3)?

More specifically, I cannot find information about the type of lesions in Task 1 and Task 2. I understand that the purpose is not to do classification, but I would like to run some experiments knowing both the segmentation mask + lesion class.

Hi @margiki, as images are stored within the isic archive gallery you should be able to get metadata for those images at https://www.isic-archive.com/#!/topWithHeader/onlyHeaderTop/gallery
Of you are able to wait a little longer to run experiments I might be able to provide/publish a significantly higher number of segmentation masks in the near future.

Thanks a lot for pointing the gallery @philipp.tschandl, it contains the metadata I needed!

However, I cannot retrieve the data via the API. For example, I am trying to retrieve the image below by calling the API https://isic-archive.com/api/v1/image/5a2ecc5e1165975c945942a2, and it returns an error “Invalid item id…”

Is there something I am missing?

It seems the Unique-ID fields aren’t the Image-IDs needed for the API. As I am not familiar with the implementation of the ISIC Gallery Front-End I won’t be able to help here, maybe @brianhelba, if he sees this, can. Until then you can select any image in the gallery and following choose to “Download all metadata in ISIC-archive” which is probably the easiest option:

1 Like

@margiki thank you for highlighting this problem in the Gallery. The Unique ID previously represented the /dataset/{id}/. Now the Unique ID corresponds to the /image/{id}/ endpoint in the API tool.