Can non-Python users contribute to the challenge?

Good morning,

Congrats for putting together an exciting challenge ! I’m almost ready to submit a contribution, but I just realized from the tutorial that Python is required. Am I missing something? There are other options out there that can be useful for machine learning (e.g. R or Julia; I’m a R user myself ;-)), and I was wondering whether there would be a way for us non-Python users to submit anyway? Does the Grand Challenge platform allow it (not sure from what I can see at https://github.com/comic/grand-challenge.org) ?

My two cents below. Just ignore it if you think my question is meaningless.

I see that the holdout dataset (test data) is not available. There could still be a simple solution. When downloading the data following https://stoic2021.grand-challenge.org/stoic-db/, we get all 10,735 patients CT scans, including those from the patients in the holdout dataset. Therefore we would just need the IDs of the patient in the holdout dataset to be able to predict using our fancy models whether they had a positive PCR and developed severe disease. Then we would just have to submit these predictions which could be confronted to the truth by the organizers. Finally our score could be sent back to us.

Cheers, and happy New Year ! Olivier

Good morning Olivier,

Great to hear that you would like to contribute to the challenge. The baseline repo is indeed Python based, but it is not necessary to stick to Python for your submission. The repo shows the input and output formats that submitted algorithms should expect and produce. You can change the Dockerfile and create a Docker container that uses any programming language you like, as long as the output and expected input adhere to these formats. Both the baseline repo and the evalutils package that it is based on are only offered in Python. So using another language than Python would require extra work.

Thank you also for thinking along with the organization of STOIC2021. We chose to completely hide the test set from participants to make the challenge as fair as possible. Furthermore, since the challenge is about sensitive data, we are unable to make the complete dataset publicly available.

Happy New Year and good luck with the challenge! Luuk

Good morning, Many thanks Luuk for the detailed response. I'll see if I understand what extra work is needed ;-) In case others are interested in R code to train a CNN on CT scans, I've written a quick and dirty post https://oliviergimenez.github.io/blog/image-classif/. I'll post an update if I manage to submit something through the challenge's platform. Cheers, Olivier