search

zhengyang-wang / 3D-Unet--Tensorflow

3D Unet for Isointense Infant Brain Image Segmentation
GNU General Public License v3.0
195 stars 97 forks source link

label data format #17

Open guava1427 opened 4 years ago

guava1427 commented 4 years ago

Hello, I'm trying to use your network. For the multi-labelled segmentation, what is the proper structure in case of the labelled data?

Your code seems to load .tfrecord (generated from .hdr by generate_tfrecord.py). I'm confused about how to make HDR files as per your intention because each slice can have multiple label binary images.

guava1427 commented 4 years ago

I found the answer in case that the labelled areas are mutually exclusive. Otherwise, how should I deal with this?