Sli2Vol: Annotate a 3D Volume from a Single Slice with Self-Supervised Learning

This repository contains the codes (in PyTorch) for the framework introduced in the following paper:

Sli2Vol: Annotate a 3D Volume from a Single Slice with Self-Supervised Learning [Paper] [Project Page]

@article{yeung2021sli2vol,
	title = {Sli2Vol: Annotate a 3D Volume from a Single Slice with Self-Supervised Learning},
	author = {Yeung, Pak-Hei and Namburete, Ana IL and Xie, Weidi},
	booktitle = {International conference on Medical Image Computing and Computer Assisted Intervention},
	pages = {69--79},
	year = {2021},
}

Contents

1. Dependencies
2. Correspondence Flow Network
3. Verification Module (TODO)

Dependencies

• Python (3.6), other versions should also work
• PyTorch (1.6), other versions should also work

Correspondence Flow Network

1. The correspondence flow network as described in the paper is coded as the class Correspondence_Flow_Net in model.py
2. It computes the affinity matrix between the input slice1_input and slice2_input and use the matrix to reconstruct slice2_reconstructed from the input slice1