Realesrgan

Please be aware that this is a pre-release. There may be bugs to be fixed, so feel free to open a new issue when you encounter one.
This pre-release implements using directory path as input and preserving alpha channel. However, due to the limitations of the BICUBIC interpolation algorithm, the alpha channel is kinda blurry around the edges.
Upscaling alpha channel with the Real-ESRGAN will be implemented in the future, as well as the traditional resampling algorithms.

This release added some more models (x2, animevideo, etc.) converted from the original Real-ESRGAN repo and fixed a bug that will produce a stripped image if an image with an alpha channel is used as input.

Please beware that you'll need to install .NET Desktop Runtime 5.0.15 or higher to run this app. Windows 10 1709 or higher is also a must if you hope to use DirectML GPU acceleration.

This release added GPU acceleration with DirectML. To enable GPU acceleration, set the device id to the GPU index you want to use. Most systems with a single GPU will have a GPU indexed 0.
Using GPU acceleration requires Windows 10 1703 or higher. On some systems, this functionality will yield a black image due to unknown issues.

edit: I also fixed an issue that would cause the app to crash if trying to process a non-square-shaped image.

edit x2: I've noticed a bug that will cause the app to produce a stripped image when processing an image with an alpha channel. A patch has been made to address this issue. I'll release 1.0.0.2 ASAP.

- Initial release.

model

Only the basic functionality is done. Batch process is not done yet.

Note that there may be huge memory consumption if your input image is large (1000x1000 can easily eat up 18 gigs of memory).

edit: This release contains a fatal bug that will cause the app to crash if process a non-square-shaped image.

🚀 Long time no see ☄️

✨ **Highlights**

✅ [add realesr-general-x4v3 and realesr-general-wdn-x4v3](https://github.com/xinntao/Real-ESRGAN/commit/b827be13a1db242ebaea1be8669c62b757bd2796). They are very tiny models for general scenes, and they may more robust. But as they are tiny models, their performance may be limited.
✅ [support denoise strength for realesr-general-x4v3](https://github.com/xinntao/Real-ESRGAN/commit/576aaddfaf1dec031cdf580924c00f4b29b9b35a). You can use the **-dn** option to adjust the denising strength. Denoise strength. 0 for weak denoise (keep noise), 1 for strong denoise ability. Only used for the realesr-general-x4v3 model.
✅ [update inference_video: support auto download](https://github.com/xinntao/Real-ESRGAN/commit/61e81d3108f9437a9d97b02fd94439e9fc191ed0). You do not need to download models explicitly. If the models are not download, they will be downloaded automatically.

✅ [support ffmpeg stream for inference_realesrgan_video](https://github.com/xinntao/Real-ESRGAN/commit/cdc14b74a54e4247581b31c66caf93ea9e0cf159)
✅ [fix colorspace bug & support multi-gpu and multi-processing](https://github.com/xinntao/Real-ESRGAN/commit/8cb9bd403e0b8206eb69780b97b35cf7aa84bd4e)
✅ [Added GPU selection feature to python inference](https://github.com/xinntao/Real-ESRGAN/commit/6b15fc693646da1f16ba38ee47cf71a285f08390)
✅ [deal with flv format](https://github.com/xinntao/Real-ESRGAN/commit/e5e79fbde32cf96720d484511fbae2272144beb6)


📢📢📢

https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth

https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth

<p align="center">
<img src="https://raw.githubusercontent.com/xinntao/Real-ESRGAN/master/assets/realesrgan_logo.png" height=150>
</p>