RAD: Region-Aware Diffusion Models for Image Inpainting (CVPR 2025)

This repository provides the official PyTorch implementation of "RAD: Region-Aware Diffusion Models for Image Inpainting".

Paper | Supplementary Materials | Supplementary Video

Requirements

• python 3.8.16, pytorch 2.0.1
• Platform: Ubuntu 22.04, CUDA 11.8

Installation

pip install -e .[torch]

Training

python examples/unconditional_image_generation\train_RAD.py --dataset_name merkol/ffhq-256 --pretrained_model_name_or_path xutongda/adm_ffhq_256x256

python examples/unconditional_image_generation\train_RAD.py --dataset_name pcuenq/lsun-bedrooms --pretrained_model_name_or_path xutongda/adm_lsun_bedroom_256x256

python examples/unconditional_image_generation\train_RAD.py --dataset_name imagenet-1k --pretrained_model_name_or_path xutongda/adm_imagenet_256x256_unconditional

Validation data structure

├── val_data_path
    ├── mask_type1
    │   ├── original
    │   │   ├── image_0001.png
    │   │   ├── image_0002.png
    │   │   └── ...
    │   └─── mask
    │       ├── mask_0001.png
    │       ├── mask_0002.png
    │       └── ...
    ├── mask_type2
    │   ├── original
    │   │   ├── image_0001.png
    │   │   ├── image_0002.png
    │   │   └── ...
    │   └─── mask
    │       ├── mask_0001.png
    │       ├── mask_0002.png
    │       └── ...
    ...

Inpainting

python examples/unconditional_image_generation/inpaint.py --val_data_path [your validation image path] --dataset_name merkol/ffhq-256 --pretrained_model_name_or_path xutongda/adm_ffhq_256x256 --resume_from_checkpoint checkpoint-300000

python examples/unconditional_image_generation/inpaint.py --val_data_path [your validation image path] --dataset_name pcuenq/lsun-bedrooms --pretrained_model_name_or_path xutongda/adm_lsun_bedroom_256x256 --resume_from_checkpoint [your checkpoint]

python examples/unconditional_image_generation/inpaint.py --val_data_path [your validation image path] --dataset_name imagenet-1k --pretrained_model_name_or_path xutongda/adm_imagenet_256x256_unconditional --resume_from_checkpoint [your checkpoint]

Checkpoint

Google Drive

Reference

@InProceedings{Kim_2025_CVPR,
    author    = {Kim, Sora and Suh, Sungho and Lee, Minsik},
    title     = {RAD: Region-Aware Diffusion Models for Image Inpainting},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {2439-2448}
}

Acknowledgements

The implementation is based on Diffusers.