search

Luffy03 / Large-Scale-Medical

[CVPR 2024 Extension] 160K volumes (42M slices) datasets, new segmentation datasets, 31M-1.2B pre-trained models, various pre-training recipes, 50+ downstream tasks implementation
Apache License 2.0
101 stars 7 forks source link

About MSD testing sets #2

Closed jsadu826 closed 2 weeks ago

jsadu826 commented 1 month ago

Dear author, I'd like to know whether the results of MSD tasks were obtained from the official testing sets or custom testing sets split from the official training sets?

Luffy03 commented 1 month ago

Hi, the results are from the splits of training sets, which are generated from 'split_final.json' of nnUNet.

Luffy03 commented 1 month ago

Hi, you can find all the json files in each repo of Downstream.

jsadu826 commented 1 month ago

Thank you, but I did not found the 'split_final.json' of nnUNet, and the data split json files of MSD tasks in Downstream seem to use the official testing set to test.

Luffy03 commented 1 month ago

Dear jsadu, the 'split_final.json' of nnUNet is generated by nnUNet preprocessing since we want to conduct fair comparisons with nnUNet. The data split json files of MSD tasks in Downstream are for offline validation not for test since we don't have the labels of test sets.