question of inference for abdomenatlas data

[cclamd]

hi ,this is the result of abdomenatlas data for runing the project , is it right ? it did't show segmentation for 9 class

Warm regards

[WenxuanChelsea]

Hi @cclamd

Thanks for your interest in our work!

Regarding "running the project," are you referring to executing inference.py in the direct inference folder? If so, the figure you attached is not correct. We recommend using the singularity or docker options we've provided for convenience.

Hope this helps!

Thanks,
Wenxuan

[cclamd]

Hi @WenxuanChelsea
yes ,i executing inference.py in the direct inference folder,and i got combined_labels.nii.gz ,after i visulize a slice from it ,it shows as the figure attached ,so should i have to using the singularity or docker? is Option 2 not correct?
Warm regards

[cclamd]

Hi @WenxuanChelsea
i want to reproducing results using abdomenatlas data of 9 class , i used the normal unet to train on the data ,and loss is from 1 to 0.05,but when inferencing ,the results is not good

Epoch=15: Training (2123 / 5195 Steps) (dice_loss=1.00000, bce_loss=0.00334):
Epoch=15: Training (2134 / 5195 Steps) (dice_loss=1.00000, bce_loss=0.00203):
Epoch=13: Training (5194 / 5195 Steps) (dice_loss=1.00000, bce_loss=0.00183):

below is the result of re pretrain ,and the bce_loss is very low