All notable changes to this project will be documented in this file.
[1.6.4] - 2022-01-25¶
Renamed GitHub Workflows from Distribution to CD and Version test to CI.
Now caching the VapourSynth installation in GitHub CI workflow.
Now directly loads a tensor from the VideoFrame data directly, without numpy as a middleman.
Reduced overall numpy use for Tensor<->VideoFrame operations.
The half parameter/options have been removed entirely and replaced with automatic infer based on input bit-depth. You must explicitly use RGBS if you want FullTensor (float32). Integer and RGBH inputs will be converted (if needed) to float16 (HalfTensor) automatically.
Removed EOL Python 3.6 from CI Workflow.
Removed unused infer_sequence method from EGVSR arch.
Removed unused options and code from frame_to_tensor.
All manual tensor deletion statements have been removed, they do not seem to help with VRAM.
The overlap reduction code per-recursion has been removed. The overlap will now always stay at the value first provided.
Fixed a big Memory leak, that I still don’t know exactly why it happened.
Fixed minimum Python version listed under Installation docs.
Enforced a VapourSynth thread count of 1 when using EGVSR. More than one should not be used during Video Models, or you will be clog VRAM.
Improved the accuracy of clamping max size value to an equation on the exact bit depth. This fixes the accuracy of RGB 27, 30, 36, and 42.
[1.6.3] - 2022-01-24¶
Recursive tiling depth is now cached per-clip, rather than per-frame.
Updated numpy to version 1.21.1.
Dropped support for Python versions older than 3.7.
Fix another regression with rejoined tensors defaulting creation on the default device.
[1.6.2] - 2022-01-24¶
Fix another regression due to incorrect overlap scaling calculation from within
[1.6.1] - 2022-01-24¶
Fix regression due to missing overlap specification to
[1.6.0] - 2022-01-24¶
Add support for EGVSR, Arch and Network.
Add support for Real-ESRGAN-v2 aka Anime Video Models (comp. vgg-style arch).
Ability to use half-precision (fp16, HalfTensor) via
halfparameter. This can help reduce VRAM.
Created tiling utilities to tile a tensor, merge tiled tensors, and automatically tile and execute recursively.
Moved the frame/numpy/tensor utility functions out of the VSGAN class and into
Renamed HISTORY to CHANGELOG, and updated changelog to be in Keep a Changelog standard.
Moved VSGAN class from
Tiling mode is now always enabled, but will only tile if you wouldn’t have otherwise had enough VRAM.
Overlap now defaults to 16.
Separated VSGAN class into two separate Network classes, ESRGAN, and EGVSR. VSGAN is no longer used and ESRGAN/EGVSR Network classes should now be imported and used instead.
runhave been renamed to
Don’t require batch in tensor_to_clip.
Make change_order False by default in frame_to_tensor, improve rest of the param defaults.
Don’t change order to (2,0,1) for ESRGAN models, was unnecessary and caused issues with Real-ESRGANv2.
Fixed support for Python versions older than 3.8.
Fixed example VapourSynth import paths casing.
Restore support for VapourSynth API 3.
Now detaches tiles from the GPU after super-resolution, to keep space for the next tile’s super-resolution.
Add support for ESRGAN+ models, Real-ESRGAN models (including 2x and 1x if pixel-shuffle was used), and A-ESRGAN models.
Add support for Newer-New-arch in ESRGAN new-to-old state dict conversion.
Rework model/arch file system structure to /models, /models/blocks and /models/ESRGAN.
Rework ESRGAN architecture as a singular class, with all ESRGAN-specific operation done within it.
Move ESRGAN-specific blocks within ESRGAN.py.
Removed some unused blocks from RRDBNet.
clipparameter of VSGAN is a VapourSynth VideoNode object (a clip).
Move RGB clip check to the constructor of VSGAN rather than
Created new sphinx documentation, replacing the old Jekyll documentation.
Added HISTORY.md file for recording history (now CHANGELOG.md).
Reword some error/warning messages, now less opinionated and more concise.
Some attributes have been renamed to be more ambiguous in the hopes more Model Architectures get supported in the future.
Fix model chaining. It now gets the correct model and model scale values for each FrameEval call.
Fixed the pytorch extra group to correctly be optional and correctly reference a dependency.
Some type-hinting has been corrected.
Added support for all RGB formats including float.
Heavily improved main model execution code.
Replace current chunk system with a seamless chunk system using overlap.
Add self-chaining system, calls can be made directly after another.
Made torch dependency optional and pointed directly to torch+cuda. This is due to conflicting kinds of torch installation methods.
.ideafolder, added to gitignore.
Only transpose C for RGB if it’s 3-channels.
Fix type annotations on Python versions older than 3.9.
Use Python version 3.9.x for Dist workflow as 3.10 is not yet supported.
Allow specification of the input array dimension order.
Add Jekyll Documentation in
Added a VSGAN Jupyter Notebook (Colab), with an Open in Colab Badge on the README.
Drop support for Python versions older than 3.6.2, due to bugs discovered in NumPy.
Replace setup.py/setuptools with Poetry.
frame_to_np, don’t reverse to BGR as it’s unnecessary.
More efficiently write an array to a VapourSynth VideoFrame.
Inherit output clip properties from input clip.
Moved README’s information to the docs.
Reworked the CD GitHub Workflow to auto-create a GitHub Release and push to PyPI.
Remove the need for plane_count, now gets it from the input frame.
Don’t define the transposes, it’s unnecessary.
Fixed a bug with frame plane access on VapourSynth API 4.
Add ability to check what the last loaded model is via
Added type-hinting across the code base as well as some doc-strings.
A heavy warning discouraging the use of your CPU as a PyTorch device was added. Ability to use your CPU was hidden but reading the warning explains how to do so.
Reduced required VapourSynth version to 48 or newer.
Remove the conversion to RGB prior to model execution. RGB is required for the Model, but let the user decide how to convert to format, what algorithm, how to deal with matrix, and so on.
Removed setuptools from dependencies.
Add a check to ensure input clip is RGB, since auto conversion was removed.
Add missing documentation on 1.1.0‘s changes to scale and such.
Added two GitHub Action workflows for CI/CD.
Moved the majority of documentation and info from the GitHub Wikis system to the README.
scalewith values taken directly from the model state.
Check that a model has been loaded before
executecan be called.
Change the RGB conversion check’s kernel to
Removed the color-space conversion implemented in 1.0.3 as it can be a lossy operation. Let the user decide how/if to convert back to the original format. E.g., what algorithm, what matrix, and so on.
Replaced unsafe assert in
RRDBNetwith an if and raise, as asserts may be removed when optimised as python byte code files.
Detect ESRGAN old/new arch models via archaic trial-and-error.
Reworked code from Functional to Object-oriented Programming.
Improve code readability, project starting to get serious.
Add ability to tile the input to reduce VRAM (does not hide seams).
VapourSynth to requirements.
Convert back to original color-space after applying the model.
Ability to select device via argument.
README file with some basic information.
Improved RGB conversion by using