search

YazhouZhu19 / RPT

[MICCAI 2023] Few-Shot Medical Image Segmentation via a Region-enhanced Prototypical Transformer
36 stars 5 forks source link

About the process of the data #5

Closed zzzyzh closed 10 months ago

zzzyzh commented 11 months ago

Thank you for your excellent work at first! Regarding the data processing, I have a question: in section 2.1 3-dimensional pixel clustering is mentioned, and in section 3 you say that all 3D scans are reformatted into 2D axial and 2D short-axis slices. So the last inputs to the model are 2D slices yes? May I ask how you performed the 3D clustering? I'm looking forward to your reply!

YazhouZhu19 commented 10 months ago

Thanks for your attention. The inputs of model are all 2D slices. We employ the supervoxel algorithm to generate pseudo-labels for performing self-supervised learning, the supervoxel method is from the ADNet by Hansen et al. You can get details from their repository: https://github.com/sha168/ADNet

All of pre-processed data and supervoxels have been uploaded.