v0.4.0

What's Changed

• VapourSynth API4 support with API3 compatibility
• getnative() visualization enhancements by @LittlePox in #27
• support using lexpr in ColorBarsHD by @AkarinVS in #33
• support txt output in getnative by @sinsanction in #34
• getnative: fix missing spline64 support by @bodayw in #35
• getnative(): Change the interface and add some features. by @SaltyChiang in #40
• getnative(): Control the intervals for multi heights mode. by @SaltyChiang in #41
• Performance optimizations using akarin.Expr()
• dfttestMC(): Fix "mdg=True". Thanks to @Ryuukishi and @kgrabs in #28
• DeFilter(): Add argument 'step_size' to prevent divergence

Known issues

• FixTelecinedFades() may suffer from accuracy error and generate artifacts #39

Full Changelog: v0.3.0...v0.4.0

v0.3.0

New function (decorator) - @multi_scale()

One of my favorite function of this release, though I do not find it useful now. I implemented it after reading Steve Eddins's series of blogs on multiresolution pyramid. The idea first came to my mind after I read Local Laplacian Filters: Edge-aware Image Processing with a Laplacian Pyramid by Paris et al., Communications of ACM, 2015.

New IQA metric - MDSI()

A competitive FR-IQA. Written for Zopti.

CuPy examples

Bilateral, Dpid, KNLMeansCL, SigmaFilter and super-xbr are implemented. Good place for me to practice.

Other new functions

• sbr_detail() (muvsfunc_misc.py): sbr() inspired detail detection algorithm.
• fade() (muvsfunc_misc.py): Implementation of fade-in/out effect.
• gaussian() (muvsfunc_numpy.py): DCT-based gaussian convolution.
• MaskedLimitFilter(): Masked limit fIlter.
• PoissonMaskedMerge() (muvsfunc_numpy.py): Seamless cloning based on poisson editing.
• avg_decimate(): Averaging-based decimation filter, proposed by @IFeelBloated.
• YAHRmask(): #22
• Cdeblend(): #23
• S_BoxFilter(): Edge-preserving technique for image filtering, CVPR2019 (oral).
• VFRSplice(): Functional and fractions-based re-implementation.
• resize() (resize.py): Experimental wrapper function for vszimg resizer in a fmtconv-like API.

Maintenance

• super_resolution(): asymmetric padding & cropping, bug fixes.
• Correct variance calculation in LocalStatistics(). @stgn
• Update and add examples.
• Replace std.Interleave() with std.DuplicateFrames() in RandomInterleave().
• Fix error when thr and thrc are different in GradFun3(). @LittlePox
• Eliminate redundant function call in GradFun3().
• Cosmetics.

Type Hinting

Thanks to mypy, I do find several issues and they have been fixed.

v0.2.0

New features - MXNet support using vs_mxnet@kice

• Introduce muvsfunc.super_resolution as an alternative to muvsfunc_numpy.super_resolution of using pre-trained image processing models in MXNet. See example.

New features - TensorFlow support

• muvsfunc_numpy.super_resolution can load TensorFlow models now.

Maintenance

• Add parameter pad_mode in muvsfunc_numpy.super_resolution.

• Remove parameters is_caffe_model, normalize_mean, normalize_std, dynamic_range in muvsfunc_numpy.super_resolution.

• muvsfunc_numpy.super_resolution now use NumPy for data processing.

• Introduce a relative limiting in muvsfunc.SeeSaw. @Helenerineium

• Gamma correction can be used in muvsfunc.firniture and muvsfunc.SSIM_downsample. @Helenerineium

• Prefer core.resize.* over core.fmtc.resample. @Helenerineium

• Replace some core.rgvs.RemoveGrain with core.std.Convolution. @Helenerineium

• Cosmetics

v0.1.0

• The main script muvsfunc.py can be considered as stable. I will keep maintaining it, but I won't change it too much in the foreseeable future.
• Everything except the main script might be changed frequently, but I will only make new release when I make important changes to the main script.