Multimodal Multi-Head Convolutional Attention with Various Kernel Sizes for Medical Image Super-Resolution (WACV 2023) - Official Repo
Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Andreea-Iuliana Miron, Olivian Savencu, Nicolae-Catalin Ristea, Nicolae Verga and Fahad Shahbaz Khan.
π This is the official repository of the "Multimodal Multi-Head Convolutional Attention with Various Kernel Sizes for Medical Image Super-Resolution" paper accepted at WACV 2023.
π Arxiv Link: https://arxiv.org/abs/2204.04218
We propose a novel multimodal multi-head convolutional attention module for super-resolution. MHCA is a spatial-channel attention module that can be integrated into any neural network at any layer. We are also the first to perform medical image superresolution using a multimodal low-resolution input.
The present code is released under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.
We release the MHCA building block.
π We obtained new SOTA results on T2w modality on the IXI data set for the scaling factor of 2x and 4x.
π€© We release the pretrained models. Check EDSR folder and try out our models.
Method | Scale | PSNR/SSIM |
EDSR + MCHA | 2x | 40.11/0.9871 |
EDSR + MMCHA | 2x | 40.28/0.9874 |
EDSR + MCHA | 4x | 32.15/0.9418 |
EDSR + MMCHA | 4x | 32.51/0.9452 |
Please follow the instructions in Install.md.
Please cite our work if you use any material released in this repository.
@inproceedings{Georgescu-WACV-2023,
title="{Multimodal Multi-Head Convolutional Attention with Various Kernel Sizes for Medical Image Super-Resolution}",
author={Georgescu, Mariana-Iuliana and Ionescu, Radu Tudor and Miron, Andreea-Iuliana and Savencu, Olivian and Ristea, Nicolae-Catalin and Verga, Nicolae and Khan, Fahad Shahbaz},
booktitle={Proceedings of WACV},
year={2023},
publisher={IEEE}
}