This work combines Image-to-Image and Class-Guided denoising diffusion probabilistic models, and uses the JUMP-Target-2 dataset.
This is the code for the class-guided image-to-image denoising diffusion probabilistic model. In our study we train our model on a real-world dataset of microscopy images used for drug discovery, with and without incorporating metadata labels. By exploring the properties of image-to-image diffusion with relevant labels, we show that classguided image-to-image diffusion can improve the meaningful content of the reconstructed images and outperform the unguided model in useful downstream tasks.
Training
The backbone of this implementation is branched from Janspiry/Palette-Image-to-Image-Diffusion-Models and is trained in the same way. We provide our .json file "Target2.json" which contains the hyperparameters used to train our models. This will require editing to suit the requirements of your file structures and datasets.
python run.py -p train -c config/Target2.jsonTesting/inference
python run.py -p test -c config/Target2.jsonIncorporating labels:
AdaGN
description coming soon
CG
description coming soon
This work is built upon the following projects, and uses a large amount of their code:
This repository is released under the Apache 2.0 license as found in the LICENSE file.
If you find this work useful, please consider citing our paper:
@misc{https://doi.org/10.48550/arxiv.2303.08863,
doi = {10.48550/ARXIV.2303.08863},
url = {https://arxiv.org/abs/2303.08863},
author = {Cross-Zamirski, Jan Oscar and Anand, Praveen and Williams, Guy and Mouchet, Elizabeth and Wang, Yinhai and Schönlieb, Carola-Bibiane},
title = {Class-Guided Image-to-Image Diffusion: Cell Painting from Brightfield Images with Class Labels},
publisher = {arXiv},
year = {2023},
}