Prerequisites:
- vapoursynth
- vs-kernels (
pip install vskernels) - mvsfunc (
pip install git+https://github.com/HomeOfVapourSynthEvolution/mvsfunc)
This is a tool to determine which scenes in a show are descalable. If a show has multiple kernels, this tool makes it easy to choose how to descale each scene. The tool creates ranges that can be fed to jvsfunc.rfs:
from descale_analysis import test_descale_error, get_descale_ranges
from vskernels import Bilinear, Bicubic, Lanczos
kernelres0 = dict(width=1280, height=720, kernel=Bicubic(0, 0.5))
kernelres1 = dict(width=1280, height=720, fractional=719.8, kernel=Bicubic(0, 1))
kernelres2 = dict(width=1280, height=720, kernel=Bilinear(), src_top=0.2, src_height=719.6, src_left=0.2, src_width=1279.6)
ranges_list = get_descale_ranges(src, [kernelres0, kernelres1, kernelres2], "showtitle_epnum", avg_error_thr=0.015, ind_error_thr=0.02)
kernelres0_descalable_ranges = ranges_list[0]
kernelres1_descalable_ranges = ranges_list[1]
kernelres2_descalable_ranges = ranges_list[2]Note the avg_error_thr and ind_error_thr parameters in get_descale_ranges. Those are error thresholds that will reject a scene's descalability if they are exceeded. avg_error_thr is an error threshold for the scene, and ind_error_thr is an error threshold for each individual frame. The default values are extremely conservative and should be changed to fit the project.
The tool will pick the descale kernel/res with the lowest error if more than one of them fall under the error thresholds. It won't return multiple possible kernels for the same scene.
You can use test_descale_error to determine good values for these thresholds:
kernelres0 = dict(width=1280, height=720, fractional=719.8, kernel=Bicubic(0, 0.5))
test = test_descale_error(jpn, kernelres0)
test.set_output()The error values will appear in the top left of the frame.
As the first code example indicates, you can feed the function a "fractional" number (e.g. from getfnative) or a set of src_* values. You can't use both, since src_* values are derived from the fractional value.
You can use this tool to detect descalable scenes in single-kernel shows:
kernelres = dict(width=1280, height=720, kernel=Bicubic(0, 0.5))
ranges_list = get_descale_ranges(src, [kernelres], "showtitle_epnum", avg_error_thr=0.015, ind_error_thr=0.02)
descalable_ranges = ranges_list[0]