Releases: itmo153277/JoshUpscale
v1.1.0-ps2 (experimental)
Experimental build for PS2 (4:3 resolution)
For OBS 30.2+
For best quality I recommend using bicubic downscaler
v1.1.0
Changelog
Model
- More small details in upscaled image
- Better performance in INT8 mode
- Much better image quality in INT8 mode
Core
- Support for the new model
OBS Plugin
- Add pre-scale frame cropping
- Fix builds for OBS < 30.0
- Set default quantization to INT8
AviSynth Plugin
- Set default quantization to INT8
OBS Plugin
OBS Studio 30.2: obs-joshupscale-installer-30.2.exe
OBS Studio 29.1-30.1: obs-joshupscale-installer-29.1.exe
Statically linked with yaml-cpp 0.8.0 and boost-iostreams 1.85.0.
Requires NVIDIA Video Effects SDK 0.7.2
AviSynth Plugin
Statically linked with CUDA 11.8 runtime, yaml-cpp 0.8.0, boost-iostreams 1.85.0 and zlib 1.3.1.
Requires NVIDIA TensorRT 8.6.1
Models
Default
- original model with FRVSR-like architectureFast
- default model without previous frame informationSmooth
- default model with frame blending with global scene detectionAdaptive
- default model with adaptive frame blending (64x64 window)
Full Changelog: v1.0.0...v1.1.0
v1.0.0
OBS Plugin
OBS Studio 30.2: obs-joshupscale-installer-30.2.exe
OBS Studio 29.1-30.1: obs-joshupscale-installer-29.1.exe
Statically linked with yaml-cpp 0.8.0 and boost-iostreams 1.85.0.
Requires NVIDIA Video Effects SDK 0.7.2
AviSynth Plugin
Statically linked with CUDA 11.8 runtime, yaml-cpp 0.8.0, boost-iostreams 1.85.0 and zlib 1.3.1.
Requires NVIDIA TensorRT 8.6.1
Models
Default
- original model with FRVSR-like architectureFast
- default model without previous frame informationSmooth
- default model with frame blending with global scene detectionAdaptive
- default model with adaptive frame blending (64x64 window)
Full Changelog: v1.0.0-rc3...v1.0.0
v1.0.0-rc3
Changelog
- OBS plugin stability improvements
- AviSynth plugin stability improvements
- Bugfixes
- Documentation for AviSynth plugin
Full Changelog: v1.0.0-rc2...v1.0.0-rc3
OBS Plugin
Tested with OBS studio 30.0.2
Statically linked with yaml-cpp 0.8.0 and boost-iostreams 1.85.0.
Requires NVIDIA Video Effects SDK 0.7.2
AviSynth Plugin
Statically linked with CUDA 11.8 runtime, yaml-cpp 0.8.0, boost-iostreams 1.85.0 and zlib 1.3.1.
Requires NVIDIA TensorRT 8.6.1
Models
Default
- original model with FRVSR-like architectureFast
- default model without previous frame informationSmooth
- default model with frame blending with global scene detectionAdaptive
- default model with adaptive frame blending (64x64 window)
v1.0.0-rc2
Changelog
- Improved UI for OBS plugin
- Added selection for models and quantization modes
- New model with adaptive blending
- Improved build pipeline
- OBS plugin installer
- Fixed compatibility with NVIDIA VFX in OBS
- Stability improvements
Full Changelog: v1.0.0-rc1...v1.0.0-rc2
OBS Plugin
obs-joshupscale-installer.exe
Tested with OBS studio 30.0.2
Statically linked with yaml-cpp 0.8.0 and boost-iostreams 1.85.0.
Requires NVIDIA Video Effects SDK 0.7.2
AviSynth Plugin
avisynth-plugin.zip
Statically linked with CUDA 11.8 runtime, yaml-cpp 0.8.0, boost-iostreams 1.85.0 and zlib 1.3.1.
Requires NVIDIA TensorRT 8.6.1
Models
You can find release models in models.zip
v1.0.0-rc1
Changelog
- Fix OBS colour spaces
- Enable INT8 quantization
OBS Plugin
obs-plugin-tensorrt
includes the following libraries:
- yaml-cpp 0.7.0 - static
- boost-iostreams - 1.82.0 - static
- NVIDIA CUDA 12.1 runtime - static
- NVIDIA TensorRT 8.6.1 core + builder resources
You need zlib.dll
, swscale-7.dll
and avutil-58.dll
to be installed on your system (usually included with OBS).
Installation
Copy contents to your OBS folder
AviSynth plugin
- zlib 1.3 - static
- yaml-cpp 0.7.0 - static
- boost-iostreams - 1.82.0 - static
- NVIDIA CUDA 12.1 runtime - static
You need to provide your own nvinfer.dll
or copy it from obs-plugin-tensorrt
v1.0.0-rc0
OBS Plugin
obs-plugin-tensorrt
includes the following libraries:
- yaml-cpp 0.7.0 - static
- boost-iostreams - 1.82.0 - static
- NVIDIA CUDA 12.1 runtime
- NVIDIA TensorRT 8.6.1 core + builder resources
(INT8 quantization is disabled)
Installation
Copy contents to your OBS folder
0.1.0-alpha.3 preview build
Windows 10 x86_64 only
Contains:
- GUI Launcher
- Avisynth plugin
Updater:
- Updater will update
0.1.0-alpha.1
or0.1.0-alpha.2
to0.1.0-alpha.3
.
Dependencies:
- Visual C++ 2019 Redistributable (downloadable)
- NVIDIA CUDA Toolkit 11.0 (downloadable)
- NVIDIA cuDNN 8.0.5 (included)
- TensorFlow 2.4.1 (included)
- SDL 2..0.14 (included)
- SDL2_gfx 1.0.4 (included)
- FFmpeg 3.4 (included)
- wxWidgets 3.1.4 (included)
Changelog:
- HiDPI support
- Support video devices as audio source
- Add preheat before capturing video
- UI fixes
0.1.0-alpha.2 preview build
Window 10 x86_64 only
Contains:
- GUI Launcher
- Avisynth plugin
Updater:
- Updater will update
0.1.0-alpha.1
to0.1.0-alpha.2
.
Dependencies:
- Visual C++ 2019 Redistributable (downloadable)
- NVIDIA CUDA Toolkit 11.0 (downloadable)
- NVIDIA cuDNN 8.0.5 (included)
- TensorFlow 2.4.1 (included)
- SDL 2..0.14 (included)
- SDL2_gfx 1.0.4 (included)
- FFmpeg 3.4 (included)
- wxWidgets 3.1.4 (included)
Changelog:
- Bug fixes
- Ability to open files
- Ability to pass configuration options to ffmpeg
- New sync code
- Convert video to 30 fps cfr
- More debug information
- Ability to disable debug information
0.1.0-alpha.1 preview build
Window 10 x86_64 only
Contains:
- GUI Launcher
- Avisynth plugin
Dependencies:
- Visual C++ 2019 Redistributable (downloadable)
- NVIDIA CUDA Toolkit 11.0 (downloadable)
- NVIDIA cuDNN 8.0.5 (included)
- TensorFlow 2.4.1 (included)
- SDL 2..0.14 (included)
- FFmpeg 3.4 (included)
- wxWidgets 3.1.4 (included)
Changelog:
- Bug fixes
- TensorFlow update to 2.4.1
- Dataset creation scripts
- Local training scripts
- Google Colab training notebook
- Model optimization scripts
- Documentation update