Xin Jin

Thanks for the compliments! If you have any ideas or suggestions for improvement, contributions or feature requests are always welcome. Let's make this project even better together!

If you want to use the LED finetuned network, you need to add `--led` at the end of the parameters. We mentioned this issue in #6 , but it was...

The reason why it's necessary to specify `--led` is that the deployed model contains two checkpoints: one before structural reparameterization and one after. When running the `image_process.py` script, it's required...

To use architectures other than UNet, like Restormer/NAFNet. you need to specify the `-opt` options. https://github.com/Srameo/LED/blob/75f1081c77d8addeff11d0360da1746216d45499/README.md?plain=1#L198-L199 And we have detailed options for Restormer/NAFNet in commit 75f1081c77d8addeff11d0360da1746216d45499. https://github.com/Srameo/LED/blob/75f1081c77d8addeff11d0360da1746216d45499/README.md?plain=1#L186

由于近期我们组员都有些其他工作需要处理，所以训练代码最近可能不能公开。 不过Loss部分您可以先参考下这个： ```python from abc import abstractmethod from megengine import module as M import megengine.functional as F class BaseLoss(M.Module): def __init__(self, name) -> None: super().__init__() self.name = name self.losses =...

您好！感谢您对LED的关注！ 或许我可以尝试使用pydng与pyexiftool来帮助您将处理后的bayer图像重新保存为dng文件。不知道这样是否符合您的需求呢？ BTW, 如果我们的 LED 对您的项目有帮助，我们欢迎您给我们一个星标 🌟，这将是对我们进一步更新的巨大鼓励！

Probably due to some copyright restrictions, @Vandermode did not open source the source code of the ELD noise model. We reproduced ELD in the latest work of ICCV23 (the metrics...

The noise model is in this [file](https://github.com/Srameo/LED/blob/main/led/data/noise_utils/noise_generator.py), and a example of the options can be found [here](https://github.com/Srameo/LED/blob/main/options/base/noise_g/noise_g_SonyA7S2.yaml).