Tips for running your algorithm on grand challenge ¶
By: jmsmkn on Aug. 18, 2021, 1:02 p.m.
When uploading your algorithm to grand challenge please note the following:
- Please remember to set the GPU and memory requirements correctly when uploading your container. If you enable GPU support inference jobs will be assigned 1x Nvidia T4 card. CPU time is limited to 2 hours.
- It may take some time to validate your algorithm, if your container is not marked as active after 1 hour please contact grand challenge support.
- You do not need to create a new algorithm for each submission, you can update your existing algorithm by uploading a new docker container image on the container management page.
- Once your container is active please test it out before submitting it to the challenge, you can upload data using the "Try Out Algorithm" button. Please use a representative TIFF image with the resolution set (maybe there is one of these in the training data) and check that the outputs are what you expect.
- Remember that your container will not have access to the internet when executing, so all model weights must be present in your container image. You can test this locally using the
--network=none
option ofdocker run
. If you do not do this you will see errors likeTemporary failure in name resolution
in your logs. - Try your algorithm out on training data from the MIDOG GitHub repo - but not that the images there are ~30x smaller than the data in the test set so this may impact your algorithm runtime.
There are more tips in the FAQ in our developer documentation.