---
license: cc-by-4.0
pipeline_tag: image-to-image
tags:
- pytorch
- super-resolution
---

[Link to Github Release](https://github.com/Phhofm/models/releases/tag/4xHFA2k)  

# 4xHFA2k

Name: 4xHFA2k  
Author: Philip Hofmann  
Release Date: 07.05.2023  
License: CC BY 4.0  
Network: ESRGAN (RRDBNet)  
Scale: 4  
Purpose: An anime images 4x upscaling model based on musl's HFA2k dataset  
Iterations: 91,000  
batch_size: 9  
HR_size: 256  
Epoch: 318 (require iter number per epoch: 286)  
Dataset: HFA2k  
Number of train images: 2568  
OTF Training: Yes  
Pretrained_Model_G: RealESRGAN_x4plus.pth  
Training time: 36h on a  P100 GPU  

Description: 4x anime image upscaler with a bit of otf jpg compression and blur. Trained on musl's hfa2k dataset release for Anime SISR, which has been extracted from modern anime films, where the selection criteria was high SNR, no DOF and high frequency information.

Examples: https://imgsli.com/MTc2NDgx  

![Example1](https://github.com/Phhofm/models/assets/14755670/7a65af06-510f-4fe7-bdd6-cd208c09947c)
![Example2](https://github.com/Phhofm/models/assets/14755670/2b671488-55f1-440b-a00d-b49753e5b05e)
![Example3](https://github.com/Phhofm/models/assets/14755670/2d21b941-0ea4-46d4-a618-66b15b05eb47)
![Example4](https://github.com/Phhofm/models/assets/14755670/7f50a8c1-57de-43b6-8fae-8e350368e5f8)
![Example5](https://github.com/Phhofm/models/assets/14755670/a76bfb26-1b95-4492-92ec-e7feda98a70a)