PreRelease-1.6.4-rev2

All from the Initial Pre-Release +

New:

• Benchmark will now have a built in counter to signify how many tests are left.
• Benchmark will now print and store the version of TAS you are using.
• Added the new update method to the AE GUI.

Removed:

• Rife 4.14 ncnn, directml, tensorrt and cuda.

Full Changelog: v1.6.3...PreRelease-1.6.4-rev2

Pre-Release[1.6.4] - Faster NCNN

New:

• Auto-Update: The script will now check if there's a new release for TAS and allow the user to auto-update. To turn this off you can set --update to 0.
• Dynamic NCNN Models: The Rife NCNN Models have been decoupled from TAS and now will be downloaded on runtime. This is compatible with the --offline mode.
• New NCNN Interpolation Models: Rife 4.16-lite-ncnn as well as Rife 4.15-lite-ncnn have been added to the script.

Improvements:

• With the help of TNTWise we have managed to improve the performance of Rife-NCNN by up to 15%.

Removed:

• Rife 4.14 NCNN ( Will be followed by Rife 4.14 Cuda and DirectML ) in favor of Rife4.15.

Fixes:

• Rife TensorRT Ensemble models were not properly downloaded and executed.

Release [1.6.3] - TensorRT

New:

• TensorRT Backend: The script now includes TensorRT Acceleration, although it's still a work in progress. Certain workflows such as Shufflecugan-TensorRT, Span-TensorRT, and Rife 4.6-TensorRT have demonstrated significant speed improvements.
• New Upscaling Models: ShuffleCugan-TensorRT, Compact/UltraCompact/SuperUltraCompact - TensorRT, Span-TensorRT.
• New Interpolation Models: Rife v: 4.6/4.14/4.15 - TensorRT

Improved:

• CUDA Upscaling and Interpolation: Enhanced memory management has led to minor improvements in inference speed.
• Testing Methodology V3 for benchmark.py: Added new TensorRT options and optimized slower processes for faster benchmarking.
• Enhanced logging and printing functionalities with more detailed information.
• Integrated all new options into the AE GUI.

Known Issues:

• GUI.exe is non-functional.

Notes:

• In-depth Benchmarks available here.
• It's worth noting that TensorRT is still in its early stages, and I am exploring further ways to improve inference speeds. Currently, engine building is relatively straightforward with minimal variation, considering possible overlaps. All Upscaling engines support dynamic shapes with sizes between 8x8 and 1920x1080, while Interpolation sees a slight boost to accommodate resolutions up to 3840x2160.
• Future releases will likely include support for Denoise TensorRT.

Full Changelog: v1.6.2...v1.6.3

Pre-Release-1.6.3 - TensorRT Upgraded

Pre-Release-1.6.3 - TensorRT,

Testing functionality, if it works I will push to new releases.

Release [1.6.2]

New Features:

• YT-DLP Resolution Chooser: Now, select your desired resolution directly from the terminal with ease.
• Benchmark.exe: Introducing a new benchmark tool, Benchmark.exe, built upon the functionality of benchmark.py. This tool aims to enhance and streamline the benchmarking process for TAS.

Removed Features:

• Unnecessary Padding for CUDA Upscaling: Padding for CUDA Upscaling has been eliminated as Spandrel now handles it entirely.
• --ytdlp_quality Option: This option is no longer necessary and has been removed.

Fixed / Improved:

• Enhanced Input Handling: Improved handling of input with unrecognized characters like " ? ".
• Increased CRF Preset: The preset CRF for the set --encode_method has been bumped up. This adjustment aims to reduce file sizes with minimal impact on perceived quality.

Full Changelog: GitHub

Release [1.6.1] - Hotfix Release

Changes:

• NCNN: In response to user feedback regarding issues with DirectML, NCNN has been reintroduced with a limited feature set to cater to a broader user base.

• ShuffleCugan: Removed Shufflecugan-DirectML due to operational issues within the model that were not being addressed by onnxruntime. Investigating a solution for future releases.

Improvements:

• CUDA: Enhanced upscaling code for improved performance and reduced clutter.

• HelpTips: Updated readme.txt and rectified outdated text in the AE GUI for better user guidance.

Known Issues:

• GUI: Current Graphical User Interface functionality is impaired due to unidentified reasons. Efforts are directed towards developing the 2nd iteration of the GUI, targeted for release in version 1.7.0 to address this issue and enhance user experience.

Release [1.6.0] - DirectML Upscaling & Interpolation

New Features

• DirectML Backend: A new backend for upscaling and interpolation. Personal tests show approximately 20% faster inference speeds over NCNN in interpolation and anywhere from 2x-4x for upscaling over NCNN. In some cases, it seems to match or even surpass CUDA.
• Offline Mode: A new --offline flag allows the user to download all available models in one go, making the script fully offline (with the exception of YT-DLP and Depth Maps).
• Data Sharing: A new --consent flag allows users to opt-in to share select data from log.txt to a server, mainly meant for improving the script in edge case situations.
• Rife 4.15-lite: This version has been added.
• New Models: The following models have been added: "compact-directml", "ultracompact-directml", "superultracompact-directml", "span-directml", "cugan-directml", "shufflecugan-directml"

Improvements

• Performance: This release includes several performance improvements, primarily for interpolation with the introduction of a slightly better caching functionality. Interpolation speeds for Rife-Cuda should have increased by roughly 5% over previous releases.
• Code Structure: The structure of the code has been slightly adjusted.
• Logging and Printing: Logging and printing functionalities have been improved.
• Download Functionality: Switched from wget to requests + tqdm to bypass certain errors some users were running into.
• File Size: The total file size of TAS has been reduced from 1.5.0 to 1.6.0 by approximately 300MB.
• GUI: The AE GUI now saves the previous set settings across launches. The backend and layout of the AE GUI have also been improved.
• Input Handling: There is likely better handling of edge case inputs with uncommon characters.
• Models: New Cugan Model.

Removed

• NCNN Backend: The NCNN backend has been removed due to its slower speed compared to DirectML and its large space requirement.
• Removed Models: Removed Cugan's default models in favor of phhofm trained model for better results.
• Removed flags: Removed --cugan_kind for the above reason.

Fixed

• Issue #20

Benchmarks

13700k Undervolted + Overclocked, 3090 Overclocked, 32GB DDR5 6800Mhz, TAS 1.5.0/1.6.0, Nilas.

TAS 1.6.0

TAS 1.5.0

Ty to Renarchi for the thumbnail.

Release [1.5.0]

Now with a GUI!!

New:

• The script now comes bundled with a Graphical User Interface ( GUI ), it's still in Alpha and there's still a ton of changes that I would like to make to it but this is just the beginning.
• New Interpolation AIs, Rife4.15, Rife4.16-lite, Rife4.15-NCNN.
• New Denoising options, Span, DPIR, SCUNet and NAFNet, feel free to play around with them and let me know how it works out.

Improvements:

• Upscaling Speed has been increased by 10-15%.
• Interpolation through Pytorch has been increased by 10%.
• Interpolation through NCNN has been increased by 5-10%.
• The script should now allow for custom -pix_fmt with the --custom_encoder option. #16

Fixes:

• AutoCut wasn't working within the Script UI.
• Fixed: #19
• Fixed YT:DLP issues #14

Thanks to Renarchi for the thumbnail.

Full Changelog: v1.4.8...v1.5.0

Release [1.4.8]

New:

• Added Rife 4.15-ncnn, use option: --interpolate_method rife4.15-ncnn
• Added Denoise option to the After Effects script, you can utilize SPAN, SCUNet, NAFNet through SPANDREL
• Added a new Check For Update button to the After Effects Script, you can now periodically check for new versions of the script directly from After Effects.
• Removed Rife 4.14-lite and Rife 4.14-lite-ncnn due to the fact that the inference speed wasn't any faster than Rife 4.14.

Fixed:

• Rife-NCNN missing module issues.

Improved:

• Improved Rife-NCNN performance by 5-10%
• Improved After Effects Script Code.

SideNote:

• There will be no more releases until 1.5.0 which will contain an Alpha Release of TAS' Graphical User Interface.

Full Changelog: v1.4.6...v1.4.8

Release [1.4.6]

New:

• Switched NCNN Backend to TNTWise's Universal NCNN Implementation
• Added ShuffleCugan-NCNN.
• Added Rife 4.15, based on user reports it seems to better than Rife 4.14.
• NCNN performance should have been slightly improved
• Span-NCNN now also has a 4x model.

Fixed:

• NCNN Import issues.

Improved:

• Better representation for cugan's denoising methods for NCNN.
• Cleaned up some of the existing code.
• Removed unnecessary extra packages, now everything is in one place, again, thanks to TNTWise.

Full Changelog: v1.4.5...v1.4.6