search

cuttle-fish-my / GM-Guided-DG

Official Implementation of ML4H2023 paper: Gradient-Map-Guided Adaptive Domain Generalization for Cross Modality MRI Segmentation
MIT License
4 stars 0 forks source link

Different Dice score for BraTS2018 dataset #1

Closed DTA-lab closed 7 months ago

DTA-lab commented 7 months ago

Hey !! i have run the code with BraTS2018 dataset, the dice score is low than you claimed, i will attach the picture for reference. The dice score that you have claimed (T2-T1) is 0.6793 but the dice score i got (T2-T1) is 0.5554 and same for the others you can check the attached file. Looking forward for your response

(https://github.com/cuttle-fish-my/GM-Guided-DG/assets/153801064/c0809caf-6b13-4341-9724-ae775b5690b7)

DTA-lab commented 7 months ago

Screenshot from 2024-02-23 15-06-04

cuttle-fish-my commented 7 months ago

Thanks for your interest! Did you use the pre-trained model provided in google drive or the one trained from scratch? If you trained from scratch, what train config did you used?

DTA-lab commented 7 months ago

Thanks for your interest! Did you use the pre-trained model provided in google drive or the one trained from scratch? If you trained from scratch, what train config did you used?

I have trained from scratch, and i have use the same configuration as in the git hub

DTA-lab commented 7 months ago

Thanks for your interest! Did you use the pre-trained model provided in google drive or the one trained from scratch? If you trained from scratch, what train config did you used?

I have trained from scratch, and i have use the same configuration as in the git hub

config :

TRAIN_FLAGS="--data_dir ../datasets/BraTS_2018_flair2t1/train --use_fp16 True --save_dir ../saved_models/BraTS/flair/GMGDG --lr 1e-4 --batch_size 24 --save_interval 1000 --lr_anneal_steps 10000 --modality source --input_mode magnitude --in_channels 3 --heavy_aug True --norm_type BN"

python ../src/Unet_train.py $TRAIN_FLAGS

cuttle-fish-my commented 7 months ago

Screenshot from 2024-02-23 15-06-04

Ohhh! I see, it seems I forgot to set the TTA config in the test scripts, which leads to TTA mode: None in your screenshot, I've fixed this in 72ee40f! Thanks for your feedback!