search

xy0806 / miccai17-mmwhs-hybrid

If the code is helpful for your work, please cite our paper "Hybrid Loss Guided Convolutional Networks for Whole Heart Parsing" in STACOM Workshop of MICCAI 2017.
82 stars 25 forks source link

why Transfer Learning works #8

Open linmeon opened 5 years ago

linmeon commented 5 years ago

I have read your paper ,the loss func is very impressive,but why C3D works well in your project?

xy0806 commented 5 years ago

thanks for your interest in our work. For C3D, i think you can read the original paper about video recognition in natural tasks. C3D is trained with videos, which is inherent in 3D form or temporal form. It is similar to the volumetric data. C3D provides a better initilization for our model and improves the generalization ability under limited medical training data.

linmeon commented 5 years ago

i try use your code make pelvic tumor segmentation, I just need to make binary classfication, is tumor or not tumor ,so my input and output channel should remove or 1 , It should look like (1,64,64,64,1) or (1,64,64,64) ? what's defferent between them?