tabassumnova/retinal_image_to_image_translation_project

Image to Image translation between two retinal modalities using CycleGAN and Segmentator models

Retinal_Image_translation_project

Digital Fundus Image (DFI) and Fluorescein Angiography (FA) are two popular modalities of choice for retinal diagnostic. In this project, I used CycleGAN with Shape-consistency loss proposed by Zhang et al.[1] to perform image-to-image translation between these two modalities.

Requirements

• torch==1.9.0
• numpy==1.21.2
• opencv-python==4.5.3.56
• imageio==2.9.0
• albumentations==1.0.3
• gdown==3.13.0
• glob==3.5
• tqdm==4.62.2
• torchvision>=0.2.1
• dominate>=2.3.1
• visdom>=0.1.8.3

Result after Segmentation

For segmentator model, I used the implementation of [3]

Result of Image-to-Image translation

References

1. https://arxiv.org/abs/1802.09655
2. https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix
3. https://github.com/srijarkoroy/segRetino