420.69 Cupscale Unofficial Version Rev3 Hotfix

This is my and Dave's Unofficial Version of Cupscale

This is a Hotfix Release

Requires 420.69 Cupscale Unofficial Version Rev3

Fixes

• Updated Interpolation Syntax 9e605ff (Should Fix #4)
• Fix RealESRGAN (NCNN) showing an error message in logs if auto is set c56a679
• Fix crashing on locales that have a comma decimal symbol 78639d5
• PTH2NCNN now uses FP16 by default fb06093

How to install

• Download this hotfix
• Extract the contents of the hotfix to 420.69 Cupscale Unofficial Version Rev3
• Replace Existing Files

420.69 Cupscale Unofficial Version Rev3

This is my and Dave's Unofficial Version of Cupscale

Updates

• Dark Mode Title Bar
  Using DWMWA_USE_IMMERSIVE_DARK_MODE. It's possible to get dark mode window frames on Windows 10 and above.
  (Untested with Windows 8 and below)

• Updated PTH2NCNN (DrPleaseRespect/cupscale-backend)
  PTH2NCNN is completely replaced with chaiNNer's TorchToNCNN Converter modified by me to work with Cupscale

• Reworked NCNN Model Scale Detection (DrPleaseRespect/cupscale-backend)
  Now parses the actual model instead of relying on the filename. Uses chaiNNer's NCNN classes to achieve this.

• Removed Python Environment Management
  You are now unable to uninstall the embedded python. The reason for this is because the new PTH2NCNN and Model Scale Detection now runs on the embedded python.. This also means that using System Python now requires installing two new packages (onnx, onnxoptimizer, {protobuf included with onnx}). I will be adding checks for these in the Dependency Checker for Rev4

• Changed License from MIT to GPL3
  The reason for this change is due to chaiNNer using GPL3.

• Using Compositions now Properly Sorts

• Fixed MozJpeg Not Post-Processing

• If an RTX Graphics Card is detected, FP16 for PyTorch will be set as default.

• Fixed ESRGAN (Pytorch) not honoring Alpha Settings

What version to use?

• cpuonly = AMD (or if you don't want to use PyTorch)
• Turing = NVIDIA Maxwell - Turing (RTX 20 Series and Earlier)
• Ampere = NVIDIA >= Ampere (RTX 30 Series and Higher)

Notes

cpuonly

You still need to enable CPU Mode in the Settings if you are going to use PyTorch without an NVIDIA GPU.

420.69 Cupscale Unofficial Version Rev2

This is my and Dave's Unofficial Version of Cupscale

• Updated ESRGAN PyTorch (DrPleaseRespect/ESRGAN)
  It should now be compatible with Variations of the ESRGAN (RRDB) architecture, including normal ESRGAN, ESRGAN+, BSRGAN, RealSR, SPSR, Real-ESRGAN, as well as Real-ESRGANv2 (SRVGG) architecture

420.69 Cupscale Unofficial Version Rev1

This is my and Dave's Unofficial Version of Cupscale

This contains some fixes and ESRGAN+ Support and an Updated Version of RealESRGAN