search

xmengli / H-DenseUNet

TMI 2018. H-DenseUNet: Hybrid Densely Connected UNet for Liver and Tumor Segmentation from CT Volumes
526 stars 161 forks source link

Loss function #22

Open thienquang199x opened 5 years ago

thienquang199x commented 5 years ago

What is "0.78,0.65,8.57" in concat? Thanks

xmengli commented 5 years ago

weight for background, liver, and tumor.

thienquang199x commented 5 years ago

How to calculate it ? Thanks

xmengli commented 5 years ago

Call the dataloader for many iterations, then check the voxel number of the background, liver and tumor respectively.

xmengli commented 5 years ago

Call the dataloader for many iterations, then check the voxel number of the background, liver and tumor respectively.

On Wed, Apr 17, 2019 at 10:46 AM thienquang199x notifications@github.com wrote:

How to calculate it ? Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/xmengli999/H-DenseUNet/issues/22#issuecomment-483916246, or mute the thread https://github.com/notifications/unsubscribe-auth/ASRsyBn9vrRsn3JX50qlLxh8jVSkTlsdks5vhosLgaJpZM4c0C-J .

thienquang199x commented 5 years ago

Thank you very much!

thienquang199x commented 5 years ago

Why weight of tumor great than liver and background ?

xmengli commented 5 years ago

When you call dataloader for many iterations, you can find the voxel amount of tumor is small.

thienquang199x commented 5 years ago

How to calculate weight ? Tumor/sum(background,liver,tumor) ?

xmengli commented 5 years ago

Yes, the weight is calculated by Tumor/sum(background,liver,tumor) during the training stages.