search

vuptran / cardiac-segmentation

Convolutional Neural Networks for Cardiac Segmentation
107 stars 41 forks source link

how did you perform multi scale center cropping #11

Open sanajabbarweb opened 5 years ago

sanajabbarweb commented 5 years ago

hi you have mentioned in your paper in data preparation section,'' we choose multi resolution approach to crop ROI at multiple size that wholly contain ventricles." where had you performed this in your code, if not how i can do this multi scale cropping of ROI.

vuptran commented 5 years ago

Multi-scale center cropping was not performed in this code. But if you want to perform multiple crops, you can specify a different size here: https://github.com/vuptran/cardiac-segmentation/blob/master/train_sunnybrook.py#L98

sanajabbarweb commented 5 years ago

thanks for your response. kindly would you elaborate that what is meant by multi crops of ROI. as per my understanding it will increase the size of training data. can you please refer that paper from where you have read this

vuptran commented 5 years ago

In my old implementation described in the paper, I increased the size of training data by cropping the center ROI with multiple crop sizes. So I doubled the size of training data by using two different crop sizes. However, in this Keras implementation I only used one crop size and was still able to get comparable performance with the old implementation.

Halleyawoo commented 4 years ago

Hello！ Could you tell me how to realize the visualization of segmentation results