Comparisons among different anime models

English | 简体中文

Update News

• 2022/04/24: Release AnimeVideo-v3. We have made the following improvements:
  • better naturalness
  • Fewer artifacts
  • more faithful to the original colors
  • better texture restoration
  • better background restoration

Comparisons

We have compared our RealESRGAN-AnimeVideo-v3 with the following methods. Our RealESRGAN-AnimeVideo-v3 can achieve better results with faster inference speed.

• waifu2x with the hyperparameters: tile=0, noiselevel=2
• Real-CUGAN: we use the 20220227 version, the hyperparameters are: cache_mode=0, tile=0, alpha=1.
• our RealESRGAN-AnimeVideo-v3

Results

You may need to zoom in for comparing details, or click the image to see in the full size. Please note that the images in the table below are the resized and cropped patches from the original images, you can download the original inputs and outputs from Google Drive .

More natural results, better background restoration | Input | waifu2x | Real-CUGAN | RealESRGAN AnimeVideo-v3 | | :---: | :---: | :---: | :---: | | | | | | | | | | | | | | | |

Fewer artifacts, better detailed textures | Input | waifu2x | Real-CUGAN | RealESRGAN AnimeVideo-v3 | | :---: | :---: | :---: | :---: | | | | | | | | | | | | | | | | | | | | |

Other better results | Input | waifu2x | Real-CUGAN | RealESRGAN AnimeVideo-v3 | | :---: | :---: | :---: | :---: | | | | | | | | | | | | | | | | | | | | | | | | | |

Inference Speed

PyTorch

Note that we only report the model time, and ignore the IO time.

GPU	Input Resolution	waifu2x	Real-CUGAN	RealESRGAN-AnimeVideo-v3
V100	1921 x 1080	-	3.4 fps	10.0 fps
V100	1280 x 720	-	7.2 fps	22.6 fps
V100	640 x 480	-	24.4 fps	65.9 fps

ncnn

• TODO