Updates to Baseline Model ¶
By: MJJdG on Feb. 23, 2024, 9:25 a.m.
Dear participants,
Today we have made an update to the baseline algorithm container (i.e. the model weights haven't been changed). There was an issue with our custom single-process inference function that lead to lower segmentation performance then when using the standard nnUnet inference pipeline. With the recent update to the Grand Challenge platform increasing shared memory, we were able to transition back to the nnUnet code for inference. You can find the updated process.py script here.
Next to this change we will also be uploading the model weights for the baseline architecture trained using the full training dataset with the updated masks to Zenodo, in the next 2-3 days. Retraining the multiple stages of the model takes quite some time, we apologize for the delay.
Kind regards,
The ULS23 organizers
