To what extent are we allowed to use external data? For example, can we utilize CheXpert dataset if we agree to the research use agreement?
Posted by: dongkyunk @ June 29, 2023, 2:33 p.m.As stated in the Terms and Conditions (https://codalab.lisn.upsaclay.fr/competitions/12599#learn_the_details-terms_and_conditions), you may use any non-private external data or pretrained models so long as you provide links or citations to these resources in your code and/or submission to ICCV CVAMD 2023. The only stipulation is that you cannot train on MIMIC-CXR cases that are known to be in this competition's development or test sets. So, yes, you may pretrain on CheXpert.
Please also be sure to follow *all* instructions in the Terms and Conditions very carefully. For example, every submission should be accompanied by a "code/" directory containing all necessary data loading, training, and inference code. While you will be allowed to participate in the competition on CodaLab without submitting code, all participants who fail to submit reproducible code with their submissions during the test phase will not be permitted to participate in ICCV CVAMD 2023. That is, reproducible code is a prerequisite for having your paper accepted to the workshop, regardless of placement on the leaderboard.
I will make a post shortly with detailed instructions on test phase and CVAMD paper submission shortly.
Greg Holste | CXR-LT Organizer
Posted by: gholste @ June 29, 2023, 3:03 p.m.Thanks for the update!
Posted by: dongkyunk @ June 29, 2023, 4:37 p.m.My 1TB M.2 NVMe SSD is full from only this MIMIC-CXR-LT.....
Posted by: peratham.bkk @ July 8, 2023, 6:07 a.m.I would suggest preprocessing the data to reduce size. I can fit all data (including external) in 1tb
Posted by: dongkyunk @ July 8, 2023, 6:42 p.m.Good!
Posted by: peratham.bkk @ July 9, 2023, 2:48 a.m.