Hi Team,
As a part of submission, we are required to submit the docker container.
Now since I joined the competition recently, instead of 50 training images i got 368 patient images in the training dataset.
My question to the organizers is , now do i have to submit the docker container with the code that the model structure will be further trained on new images or the new images will just be used for prediction?
Currently I have built the docker container which will only take the images for the score prediction and i have placed the trained models in it.
It would be great if you could help me in answering this.
Regards,
Akshat
Created by Akshat Shreemali akshat91 Hi @vladyorsh; the server should be working now. thanks for your patience! Thanks, I'll be waiting for news. I believe so - we were doing some testing and ran into the same issue in the past thirty minutes or so. It looks like there may be a `slurm` outage on UAB's end. We reached out to UAB who are looking into it! Thanks for response. One more question - is the server is down now? My fast lane job threw an ERROR immediately after submission two times now with no logs. Yes, this is correct. At the outset of the challenge we were allowed by the data contributor to release only 50 patients of images publicly and have the total set only on the UAB challenge processing server, but we were later permitted to release the full dataset.
Participants can either train locally, and submit a pretrained model, or they can train on the server as part of the submission (if, for example, they don't have the computational resources to run a complicated model with many epochs locally).
Best,
Robert
"When we run your container we provide the (same) training images...."
Can I clarify - does it mean that publicly available set of 368 patients and the set on which the model is trained on the server are equal? Hi there,
When we run your container we provide the (same) training images to your container, so that you can train as part of the submission process if you like.
If you submit a pretrained model that does not use the training images, that is fine too. The only additional images provided during submission are the `/test/` images.
Best,
Robert