search

IAmSuyogJadhav / 3d-mri-brain-tumor-segmentation-using-autoencoder-regularization

Keras implementation of the paper "3D MRI brain tumor segmentation using autoencoder regularization" by Myronenko A. (https://arxiv.org/abs/1810.11654).
MIT License
369 stars 113 forks source link

How much GPU memory does the model need? #11

Closed ljw-struggle closed 5 years ago

ljw-struggle commented 5 years ago

I don't have Tesla V-100. Can i run it with two GTX 1080 ti?

yf0726 commented 5 years ago

Hello,

I am afraid not. I tried a very small patch size (64,80,96, batch size=1) and had oom error.

But on 2080ti every thing goes on well (but still I need to crop the image into patch size=96, 112, 80), and finally I got 0.86 dice score on BraTS2018 and 0.87 on another dataset.

ljw-struggle commented 5 years ago

Hello,

I am afraid not. I tried a very small patch size (64,80,96, batch size=1) and had oom error.

But on 2080ti every thing goes on well (but still I need to crop the image into patch size=96, 112, 80), and finally I got 0.86 dice score on BraTS2018 and 0.87 on another dataset.

Thank you, your answer is helpful.