FAQs and deadline reminder ¶
By: sliew on Aug. 12, 2022, 4:54 p.m.
*Reminder: * The deadline for BOTH your full docker submission and the abstract/BrainLes@MICCAI registration/github link for your algorithm as submitted via this form (https://forms.gle/RBhLJFXTBa33Ko9T9) are due August 17 at midnight Pacific Time. Please note that extensions will NOT be granted for failed submissions (unless it is caused by an error on our end). Please ensure that your docker runs properly in advance of the deadline. In other words, DO NOT WAIT UNTIL AUGUST 17 TO TRY THE SANITY CHECK!
*Q. If I am on a team, can each person from the team submit a docker? *
A. No. Each team should select ONE person to submit their docker entry (otherwise this would penalize smaller teams). If your team has submitted more than one docker, only the submission ID that your team lists on the challenge form (https://forms.gle/RBhLJFXTBa33Ko9T9) will be used, and if that is not indicated, then the first submission will be used.
If you wish to submit multiple approaches, you will need to form separate teams for each approach (and a participant can only be on one team).
*Q. Is the ATLAS docker submission process similar to the ISLES’22 docker submission process? *
A. Yes! The process is extremely similar. Our ATLAS docker template is here: https://github.com/npnl/atlassampledocker and you just need to follow the instructions in the readme (modify requirements.txt, process.py, and Dockerfile), build, and export. We strongly encourage you to complete the sanity check early to ensure you understand how it works; the full submission will be essentially the same.
*Q. Is there a run-time limit on the docker? *
A. You will get up to 15 minutes per subject, but we expect most submissions to be closer to ~5 min per subject, as there is no training – only inference/evaluation, so you will just submit the model weights you have, load them, apply preprocessing, and run it through your model. The hidden test set size is n=312. Also, note that execution time is different from clock time (when it starts to when it completes), since there is a queue, so the total time from the time you submit to the time it finishes may be different from your actual run time.
Q. Can the docker use GPUs?
A. Yes – the examples we provide are built around pytorch. When you submit the algorithm, click that you will use GPU and a GPU node will be made available for your model. If you request a GPU, your container will get exclusive access to 1x NVIDIA T4 GPU.
Q. Is there a limit for the total size of the Docker?
A. The container image must be less than 15 GB uncompressed.
*Q. Are fuzzy predictions ok? *
*Q. There was an issue identified with the sanity check container as of August 11, 2022 7 am PT. A fix is being uploaded August 12, 2022 and will be announced in the related forum thread.
This FAQ will be updated as additional questions come in.