jusiro / FLAIR

FLAIR: A Foundation LAnguage-Image model of the Retina for fundus image understanding.
https://jusiro.github.io/projects/flair
Apache License 2.0
75 stars 7 forks source link

downloading link for the 05_1000x39 dataset #3

Open wk100869 opened 8 months ago

wk100869 commented 8 months ago

Hello, I am very interested in your article and have reproduced it. We mainly tested two single disease, DR and macular membrane-ERM, and performed well on DR. However, there were many missed diagnoses on the macular membrane dataset, possibly due to a small number of datasets. 04 RFMid and 05 1000x39 datasets all contains macular membrane ERM and 05 1000x39, we only used 04 The RFMid dataset, and cannot download the dataset 05_1000x39. we still want to use 05_1000x39 to test macular membrane-ERM disease. Could you send download link for 05_1000x39 dataset? We look forward to and thank you very much for your reply!

Xyporz commented 5 months ago

I also need the 05_1000x39 dataset, and I have noticed that the link to the 01_EYEPACS dataset provided in the readme file is no longer active. How can I obtain it? Looking forward to and appreciate the author's reply.

jusiro commented 4 months ago

Hi both!

Sorry for the late response.

Regarding @wk100869 comment: (i) We expect the zero-shot performance to correlate highly with the proportion of each particular disease/lesion in the pre-training dataset. We encourage you to try the few-shot adaptation for less frequent diseases. In addition, you can try to design expert text prompts describing the target disease through specific lesions or findings, which may also appear during pre-training. (ii) The link for 05_1000x39 dataset is the following: LINK.

Regarding @Xyporz comment: (i) For the 05_1000x39 dataset, see the answer above. (ii) Thanks for noticing the broken link for 01_EYEPACS! We have corrected it.

Thank you for your interest in our project!

Regards,

Julio Silva-Rodriguez

Xyporz commented 4 months ago

Thank you for your response! However, I would like to point out that the new link for 01_EYEPACS seem does not provide the ["train", "test", "val"] CSV files, which are needed for the prepare_partitions.py script.