Home

Awesome

Evaluating the feasibility of using Generative Models to generate Chest X-Ray Data

Using GANs and Stable Diffusion to generate Chest Xray data points and evaluating them using convolutional classifiers.

This repository is the official implementation of Evaluating the feasibility of using Generative Models to generate Chest X-Ray Data

How to run

Training

To train the models in the paper, open this notebook:

Chest XRay Image Synthesis The PGGAN is pre-trained and can only be used to generate samples, meanwhile the stable diffusion model is fine-tuned in the notebook and saved to huggingface. (Make a copy if you wish to make changes.)

Data Generation and Evaluation

To generate the chest x-ray data and evaluate using a classifier, open this notebook:

Data Synthesis For Image Classification This notebook is used to genarate multiple images to be used as a dataset or in addition to a dataset. (make a copy if you wish to make changes)

Processed Dataset

As mentioned in the paper, we have processed the Chest X-ray 14 dataset to prepare it for stable diffusion fine-tuning. The images can be found here

Pre-trained Models

You can download pretrained model here:

Use the data generation and evalutation notebook linked above to use any of these models yourself.

Results

Dataset

All our models and evaluations have utilised the Chest Xray14 dataset: ChestXray14

Other credits