peterzhousz / ddrm Goto Github PK

# Denoising Diffusion Restoration Models (DDRM)

[Bahjat Kawar](https://il.linkedin.com/in/bahjat-k), [Michael Elad](https://elad.cs.technion.ac.il/), [Stefano Ermon](http://cs.stanford.edu/~ermon), [Jiaming Song](http://tsong.me)

DDRM use pre-trained [DDPMs](https://hojonathanho.github.io/diffusion/) for solving general linear inverse problems. It does so efficiently and without problem-specific supervised training.

Code is coming soon...

## Running the Experiments
The code has been tested on PyTorch 1.8.

### Pretrained models
We use pretrained models from [https://github.com/openai/guided-diffusion](https://github.com/openai/guided-diffusion), [https://github.com/pesser/pytorch_diffusion](https://github.com/pesser/pytorch_diffusion) and [https://github.com/ermongroup/SDEdit](https://github.com/ermongroup/SDEdit)

We use 1,000 images from the ImageNet validation set for comparison with other methods. The list of images is taken from [https://github.com/XingangPan/deep-generative-prior/](https://github.com/XingangPan/deep-generative-prior/)

The models and datasets are placed in the `exp/` folder as follows:
```bash
<exp> # a folder named by the argument `--exp` given to main.py
├── datasets # all dataset files
│   ├── celeba # all CelebA files
│   ├── imagenet # all ImageNet files
│   ├── ood # out of distribution ImageNet images
│   ├── ood_bedroom # out of distribution bedroom images
│   ├── ood_cat # out of distribution cat images
│   └── ood_celeba # out of distribution CelebA images
├── logs # contains checkpoints and samples produced during training
│   ├── celeba
│   │   └── celeba_hq.ckpt # the checkpoint file for CelebA-HQ
│   ├── diffusion_models_converted
│   │   └── ema_diffusion_lsun_<category>_model
│   │       └── model-x.ckpt # the checkpoint file saved at the x-th training iteration
│   ├── imagenet # ImageNet checkpoint files
│   │   ├── 256x256_classifier.pt
│   │   ├── 256x256_diffusion.pt
│   │   ├── 256x256_diffusion_uncond.pt
│   │   ├── 512x512_classifier.pt
│   │   └── 512x512_diffusion.pt
├── image_samples # contains generated samples
└── imagenet_val_1k.txt # list of the 1k images used in ImageNet-1K.
```

### Sampling from the model

```
python main.py --ni --config {CONFIG}.yml --doc {DATASET} --timesteps {STEPS} --eta {ETA} --etaB {ETA_B} --deg {DEGRADATION} --sigma_0 {SIGMA_0}
```
where the following are options
- `ETA` is the eta hyperparameter in the paper. (default: `0.85`)
- `ETA_B` is the eta_b hyperparameter in the paper. (default: `1`)
- `STEPS` controls how many timesteps used in the process.
- `DEGREDATION` is the type of degredation allowed. (One of: `cs2`, `cs4`, `inp`, `inp_lolcat`, `inp_lorem`, `deno`, `deblur_uni`, `deblur_gauss`, `sr2`, `sr4`, `sr8`, `sr16`, `color`)
- `SIGMA_0` is the noise observed in y.
- `CONFIG` is the name of the config file (see `configs/` for a list)
- `DATASET` is the name of the dataset used, to determine where the checkpoint file is found.

For example, for sampling noisy 4x super resolution from the ImageNet 256x256 unconditional model using 20 steps:
```
python main.py --ni --config imagenet_256.yml --doc imagenet --timesteps 20 --eta 0.85 --etaB 1 --deg sr4 --sigma_0 0.05
```

## References and Acknowledgements
```
@article{kawar2022denoising,
  title={Denoising Diffusion Restoration Models},
  author={Kawar, Bahjat and Elad, Michael and Ermon, Stefano and Song, Jiaming},
  journal={arXiv},
  year={2022},
  month={January},
  abbr={Preprint},
  url={https://arxiv.org/abs/?}
}
```

This implementation is based on / inspired by:
- [https://github.com/hojonathanho/diffusion](https://github.com/hojonathanho/diffusion) (the DDPM TensorFlow repo),
- [https://github.com/pesser/pytorch_diffusion](https://github.com/pesser/pytorch_diffusion) (PyTorch helper that loads the DDPM model), and
- [https://github.com/ermongroup/ddim](https://github.com/ermongroup/ddim) (code structure)