The repo provides the demo code for the nonuniform deblurring method. For more efficient Langevin sampling, we provide the preconditioned gradient descent Langevin (p-LD) iteration, which is used in the demo code.

All the codes are tested on Ubuntu LTS 18.04. Pls ensure the working OS system matches the required OS version.

The following two demo files are accessible upon the request at [email protected].

Lai_nonuniform_cvpr_MCEM_pLD.py
Gopro_nonuniform_cvpr_MCEM_pLD.py

You can access the forward_operator demo file and the reconstruction of the Lai nonuniform dataset at Google driver. The quantitative metric is also provided, where the rotation is taken into consideration.

```bash
python3 Lai_nonuniform_cvpr_MCEM_pLD.py
```

```bash
python3 Gopro_nonuniform_cvpr_MCEM_pLD.py
```

If you find our work useful in your research or publication, please cite it:

@inproceedings{li2023self,
  title={Self-Supervised Blind Motion Deblurring With Deep Expectation Maximization},
  author={Li, Ji and Wang, Weixi and Nan, Yuesong and Ji, Hui},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={13986--13996},
  year={2023}
}