imranrszd/chest-xray-gan-enhancement

Chest X-ray Enhancement with a Conditional GAN

A free, static demo of my Final Year Project. A trained norm-free U-Net generator (Pix2Pix-style conditional GAN) enhances low-quality chest X-rays.

The model runs entirely in your browser via ONNX Runtime Web — there is no server. Upload a degraded X-ray (or pick an example) and it returns a sharper, denoised version.

Trained with a two-stage recipe: reconstruction pretrain (L1 + edge), then adversarial finetune (L1 + VGG perceptual + Sobel edge). See the full project repo for training, evaluation, and ablations.

Research demo only — AI-generated images must not be used for clinical diagnosis.