Awesome
Test-Time Generative Augmentation (TTGA)
This is the official repository for "Test-Time Generative Augmentation for Medical Image Segmentation"
Introduction
Test-Time Generative Augmentation (TTGA) is a novel approach to enhance medical image segmentation during test time. Instead of employing handcrafted transforms or functions on the input test image to create multiple views for test-time augmentation, this approach advocate for the utilization of an advanced domain-fine-tuned generative model, e.g., diffusion models, for test-time augmentation. Hence, by integrating the generative model into test-time augmentation, we can effectively generate multiple views of a given test sample, aligning with the content and appearance characteristics of the sample and the related local data distribution.
<img src="figs/fig-1.png">Augmentation
:sparkles: Optic Disc and Cup Segmentation
<p float="left"> <img src=figs/fundus_org.png height=150 /> <img src=figs/fundus_aug.gif height=150 /> </p>:sparkles: Polyp Segmentation
<p float="left"> <img src=figs/polyp_org.png height=150 /> <img src=figs/polyp_aug.gif height=150 /> </p>:sparkles: Skin Lesion Segmentation
<p float="left"> <img src=figs/skin_org.png height=150 /> <img src=figs/skin_aug.gif height=150 /> </p>Materials
:two_hearts: SOTA segmentation models with codes, datasets and open-source parameters. (Thanks!)
| Index | Physiology | Dataset | Paper | Code |
|---|---|---|---|---|
| 1 | Optic Disc and Cup | REFUGE20 | Segtrain | code |
| 2 | Polyp | Kvasir<br>CVC-ClinicDB<br>CVC-ColonDB<br>CVC-300<br>ETIS-LaribPolypDB | HSNet | code |
| 3 | Skin Lesion | ISIC 2017<br>ISIC 2018 | TMUnet | code |
Citing
TO-DO.