Step 1: Read this article.

Step 2: Turn on the computer.

Step 3: Download and install.

Hello, everyone! I am ma Jianchang who has been working for three days.

The Spring Festival holiday has finally come to an end, the home under the blizzard New Year’s day also can not go out of the door, idle is idle, it is better to soak in B station to review childhood animation. I forgot about work a long time ago.

Even if you watch cartoons, you’re learning about open source. In order to prove that their holidays do not give up the spirit of hard study, today to share the home for seven days research results.

01

Three red models with fixed image quality

Back in the day, the black Cat Sheriff was full of flickering, jiggling, splotchy film visuals that smelled like old morning. Nowadays, watching old cartoons in STATION B can change to HD in one second. Behind the technology that turns old animation into hd stands a universal super-resolution model for animation graphics called Real-Cugan. This project is an official development of STATION B dedicated to the restoration of station B animation quality model, now open source.

Real-cugan, short for Real Cascade U-Nets for Anime Image Super Resolution. Real-cugan supports 2x/ 3X /4x super-resolution, in which the 2x model supports four noise reduction strengths and conservative repair, and the 3x/4x model supports two noise reduction strengths and conservative repair. In addition, real-Cugan uses training code derived from a mega anime data RealESRGAN, whose structure is compatible with Waiuf2x, a previously popular hyperpartitioned model. Both models are open source, and both have taken the Internet by storm.



So where do they come from?

One day, a Japanese technologist had a flash of inspiration when he was studying a paper. He wondered if he could reduce the noise and enlarge an image, so as to make an unclear illustration picture clear. So he ran an ARTIFICIAL neural network to shrink a bunch of GAL images and place the original images together so Waifu could learn how to zoom in and stretch. Thus Waiuf2x was born.

Waiuf2x is a lossless 2d image magnification software that provides noise reduction, magnification, noise reduction and magnification. However, with the expansion of the application range, not only two-dimensional images, other images and 4K wallpaper can be used.

Unlike Waiuf2x, RealESRGAN has a luxury r&d lineup. Tencent has a research institute called “scout” and “special forces” of Tencent PCG. They always stand in the front line of Tencent’s exploration and challenge of cutting-edge technologies related to intelligent media, which is Tencent ARC Laboratory.

In September 2021, the team released an image super-resolution model called RealESRGAN that effectively eliminates ringing and overshoot artifacts in low-resolution images, while also restoring more realistic details. This is not the first one. In fact, RealESRGAN is a perfect and in-depth study of its predecessor ESRGAN, so as to be able to repair images more accurately.

However, some people may wonder, the three popular super score models are excellent, but should learn to use which one? Or is there a model that does all three?

02

Recommend a comprehensive super – segmentation model

Today, we will introduce an open source program on Gitee that has 525 stars and 17% Gitee index. It integrates Waifu2x, Real-ESrGAN, real-CuGAN, as well as SRMD, Anime4K, RIFE, CAIN, DAIN and ACNet. And can provide zoom and noise reduction (super resolution) and video fill frame (insert frame) functions for pictures, GIFs and videos.

This model is waifu2X-extension-GUI.

Open Source License:

Project the author

aaronfeng 

Project introduction

This is a very useful image magnification and denoising tool, with the help of deep convolutional neural network can enlarge the image or GIF image to the size you want, and will not blur, but will become clearer and clearer, at the same time, also support GIF GIF optimization and video frame filling/interpolation.

The project address

Gitee.com/aaronfeng07…

Model features:

1. Automatic processing of a variety of media: while supporting the processing of pictures, the automatic processing of GIF&APNG and video is realized.

2. Full image style support: a variety of built-in algorithms, whether it is two-dimensional animation or your daily photos & videos, can be clear processing.

3. Video framing (supplementary framing): After the super resolution of the video, the video can be automatically framed to improve the video smoothness.

4. Video picture analysis: accelerate video processing speed and improve video picture quality through intelligent analysis of video picture.

5. Great compatibility: Multiple engines built in, compatible with almost all modern Windows PCS.

6. Flexible multi-threaded Settings: While providing complete Settings for each engine, you can also freely adjust the number of threads to take full advantage of all the performance of the PC.

7. Multiple GPU support: All built-in engines can support multiple graphics cards to make full use of all the performance of your PC.

8. Numerous custom Settings: You can adjust numerous custom Settings within the software for an automated workflow.

9. Intelligent setting: this software can automatically adjust some software Settings according to your PC hardware information and the information of the files you want to process.

Easy to use graphical user interface.

Built-in algorithm code:

Built-in super-resolution algorithm: Waifu2x/SRMD/RealSR/Real-ESrGAN/Real-Cugan/Anime4K/ACNet Waifu2x-caffe / Waifu2x-converter / Waifu2x-ncnn-vulkan SRMD-ncnn-vulkan / RealSR-ncnn-vulkan / Anime4KCPP / SRMD-CUDA RIFE/CAIN/DAIN Built-in frame interpolation engine: rife-ncnn-vulkan / cain-ncnn-vulkan / dain-ncnn-vulkanCopy the code

Operation legend:







The Demo presentation:

2 d animation:



GIF dynamic graph:

Before

After

Download from pan.baidu.com/s/1TO4a_N2Q… Those interested can also go to gitee.com/aaronfeng07… Author warehouse, directly download and use it.