The .csv data of dc_224 isn't complete

[ivanwu0404]

Dear usuyama, I'd like to use the pill data from your download link, but I found that in the .csv file, the part of dc_224 only have 3728 data but not 5000 data, is that correct ? Or there are something I didn't notice ?Thank you.

[usuyama]

Hi @ivanwu0404 thank you for checking out this benchmark.

As you said, originally, there're 5k consumer images in total, but for some pill types, we couldn't form any front and back matches. We wanted to focus on the both-sides scenario where models/users have access to both front and back images, so those images were filtered in the dc_224 folder for the experiments.

Let me know if you have any questions.

[ivanwu0404]

Ok I realized, thanks for your reply

[ivanwu0404]

Sorry, there is another question I want to ask for, the same pills' pilltype_id in fcn_mix_weight and segmented folder is different, do you have any method using the different pilltype_id to find the same pills ?

[usuyama]

@ivanwu0404 pilltype_id should be same for same pills across the dataset, although it's possible to have some errors e.g. same pills by different manufactures have different NDC. Do you have examples where they have different pilltype_id?