citysmall / qualityscaler

Image/video deeplearning upscaler app for Windows - BRSGAN & RealSR_JPEG

Image/video deepLearning upscaler app for Windows

Credits.

BSRGAN - https://github.com/cszn/BSRGAN | https://arxiv.org/abs/2103.14006

RealSR_JPEG - https://github.com/jixiaozhong/RealSR | https://arxiv.org/pdf/2005.01996.pdf

How is made.

QualityScaler is completely written in Python, from backend to frontend. External packages are:

• AI -> Pytorch-directml
• GUI -> Tkinter / Tkdnd / Sv_ttk
• Image/video -> OpenCV / Moviepy
• Packaging -> Pyinstaller
• Miscellaneous -> Pywin32 / Win32mica / Image_slicer

Installation.

1. download the QualityScaler release .zip
2. unzip using 7zip or similar
3. execute QualityScaler.exe in the directory

Requirements

• Windows 11 / Windows 10
• RAM >= 8Gb
• Directx12 compatible GPU:
  • any AMD >= Radeon HD 7000 series
  • any Intel HD Integrated >= 4th-gen core
  • any NVIDIA >= GTX 600 series
• CPU [works without GPU, but is very slow]

Features.

• Easy to use GUI
• Image/list of images upscale
• Video upscale
• Drag&drop files [image/multiple images/video]
• Automatic image tiling and merging to avoid gpu VRAM limitation
• Different upscale factors:
  • x2 - 500x500px -> 1000x1000px
  • x3 - 500x500px -> 1500x1500px
  • x4 - 500x500px -> 2000x2000px
• Cpu and Cuda backend
• Compatible images - png, jpeg, bmp, webp, tif
• Compatible video - mp4, wemb, gif, mkv, flv, avi, mov, qt

Next steps.

• Support for Nvidia RTX 3k and 2k with cuda 11
• New Fused model (that combines the best of both models)
• New GUI with Windows 11 style
• Include audio for upscaled video
• Switch to Pytorch-directml to support all Directx12 compatible gpu (AMD, Intel, Nvidia)
• Convert to Cython core functions
  • expecting ~50% more performance
• Update libraries
  • Python 3.10 (expecting ~10% more performance)
  • Python 3.11 (expecting ~30% more performance, now in beta)

Some Example.