search

jeya-maria-jose / KiU-Net-pytorch

Official Pytorch Code of KiU-Net for Image/3D Segmentation - MICCAI 2020 (Oral), IEEE TMI
https://sites.google.com/view/kiunet/kiu-net
MIT License
357 stars 80 forks source link

for the LiTS dataset, why you # the kiunet model in net/models.py #5

Closed jiangjiaxi96 closed 3 years ago

jeya-maria-jose commented 3 years ago

The commented model is just an extended model with more layers of the kiunet model found in net/models.py. Feel free to use that if the number of parameters is not a constraint for your experiments.

jiangjiaxi20 commented 3 years ago

thank you so much! In the LiTS file. i read your paper, the # kiunet seems to be the structure you mentioned in your paper, the un-#-kiunet seems do not fit your structure in your paper at all. And in the line 656 net=segnet(training=True), should i change this to net=kiunet(training=true) in order to use the kiunet?

jeya-maria-jose commented 3 years ago

Hi, Yes, loading the # kiunet model requires more memory and takes more time to train, so un-#-kiunet is a simpler version of it which gives comparable performance. To get exact results as of paper, please use # kiunet model. Yes, change line 656 to specify the model.

jiangjiaxi96 commented 3 years ago

Thank you so much for your help!!! but I trained the model, and run the val.py. the pred.nii did not show the tumor parts. I only see the Liver contour in the segmentation results.

On Oct 28, 2020, at 1:24 AM, Jeya Maria Jose notifications@github.com wrote:

Hi, Yes, loading the # kiunet model requires more memory and takes more time to train, so un-#-kiunet is a simpler version of it which gives comparable performance. To get exact results as of paper, please use # kiunet model. Yes, change line 656 to specify the model.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jeya-maria-jose/KiU-Net-pytorch/issues/5#issuecomment-717400479, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBQOKKTKOOFBUNBHFTJBOLSM3655ANCNFSM4S7CFYGA.

jeya-maria-jose commented 3 years ago

Yes, it will not. We train the network only to detect liver in this work. We do not consider the second class (tumor).

jiangjiaxi96 commented 3 years ago

What is your best dice and jacquard score in the LiTS dataset by using the minuet

On Oct 29, 2020, at 8:20 PM, Jeya Maria Jose notifications@github.com wrote:

Yes, it will not. We train the network only to detect liver in this work. We do not consider the second class (tumor).

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jeya-maria-jose/KiU-Net-pytorch/issues/5#issuecomment-718716805, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANBQOKLXEATETZXHHFOTM5LSNFM2BANCNFSM4S7CFYGA.