search

NVIDIA / DeepLearningExamples

State-of-the-Art Deep Learning scripts organized by models - easy to train and deploy with reproducible accuracy and performance on enterprise-grade infrastructure.
12.94k stars 3.12k forks source link

training and validation loss in brats 2022 challenge #1307

Open abbas695 opened 1 year ago

abbas695 commented 1 year ago

in your brats 2022 notebook you report that you trained 5 models for 150 epochs and averaged their predictions can i know what was the range of Val loss and train loss at the end of each fold?

michal2409 commented 1 year ago

I'm sorry but I don't have this data for loss values. Please refer to this table to compare between BraTS2021 (Optimized U-Net) and BraTS 2022 (Tuned U-Net) for the mean dice score.

Screenshot 2023-06-20 at 13 45 31
abbas695 commented 1 year ago

thank you so much but does this table come from a certain paper? i know the optimized U-net values come from the "Optimized U-Net for Brain Tumor Segmentation" paper but what is Tuned U-net?

michal2409 commented 1 year ago

Tuned U-Net is the solution from BraTS 2022 notebook. Table is from our paper and will be available in proceedings (https://www.med.upenn.edu/cbica/brats)

abbas695 commented 1 year ago

Tuned U-Net is the solution from BraTS 2022 notebook. Table is from our paper and will be available in proceedings (https://www.med.upenn.edu/cbica/brats)

thank you i hope you perform well in the challenge another thing i want to ask about is in the brats22_model.py file the filters are [64, 128, 256, 512, 768, 1024, 2048] self.filters = [64, 128, 256, 512, 768, 1024, 2048][: len(strides)] but in reality the bottle neck ends with 1024 channel , so what is the purpose of 2048 ?