guided-diffusion

This is the codebase for Removing Anomalies as Noises for Industrial Defect Localization.

Installation

This repository is based on openai/guided-diffusion. Please refer to the vanilla installation.

Prepare models for training

• Pretrained 256x256 guided-diffusion model: 256x256_diffusion.pt

Prepare MVTecAD dataset

Put your MVTecAD dataset in the ./data/MVTecAD folder.

Training

For MVTecAD examples, run

bash train.sh MVTecAD $catogory 64

Testing

To test the AUROC/AUPRO performance of trained models, run

bash test.sh $category