This is an example for how to prepare a valid submission for the MICCAI L2L Challenge
It is based on the pretrained imagenet example from MIMeta and process input images and generates predictions by fine-tuning a pretrained network on the target task.
A real submission will likely be more complex and probably contain multiple python modules, but this example should give you a good starting point.
Please also note that in this example we aimed for clean and simple code, not runtime efficiency. A more efficient fine-tuning loop would allow for more steps and a larger, given the available evaluation time.
- Python 3.10
- pip
- singularity
Create a new environment with python>=3.10 and install the dependencies:
pip install -r requirements.txt
In the example_submission
directory, run download.py
to download the model. This
allow us to put the model in the container and avoid downloading it at evaluation time.
Containers will not have access to the internet on the evaluation server.
python download.py
Still in the example_submission
directory, run the following command to build the
singularity container:
singularity build --fakeroot --force example-submision.sif singularity.def
- Create an account on the L2L challenge portal and subsequently a team.
- Go to the submission page and submit
the
example-submision.sif
file.
Stefano Woerner, Bartłomiej Baranowski