Selur's Little Message Board

Full Version: Ghosting on animation
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Hi,

I have currently a problem with a bluray.
It is some animation and the same problem is also present on the DVD release (I own both).

There is some ghosting :
[Image: ghosting.jpg]

You can find a 1 minute clip with this link : https://drop.idrix.fr/r/e8RBZk52ZJ#vJ9sE...VNV7KDqJk=.

I am using the following script to do some IVTC :
Code:
import vapoursynth as vs
import havsfunc as haf
import awsmfunc as awf
from vsdehalo import fine_dehalo
import vsdeinterlace as ivtc
import vsTAAmbk as taa
import validate
import math
core = vs.core

episode = core.bs.VideoSource(source='./Clip_1min.mkv', cachemode = 4)

#episode = core.std.AssumeFPS(clip=episode, fpsnum=25, fpsden=1)

# Setting detected color matrix
episode = core.std.SetFrameProps(clip=episode, _Matrix=1)
# color transfer
if validate.transferIsInvalid(episode):
    episode = core.std.SetFrameProps(clip=episode, _Transfer=1)
# setting color primaries
if validate.primariesIsInvalid(episode):
    episode = core.std.SetFrameProps(clip=episode, _Primaries=1)

# setting color range to TV (limited) range.
episode = core.std.SetFrameProps(clip=episode, _ColorRange=1)

# making sure the detected scan type is set (detected: top field first)
episode = core.std.SetFieldBased(episode, 2)

episode = core.vivtc.VFM(clip=episode, order=0, field=2, mode=1, mchroma=True, cthresh=9, mi=80, chroma=True, blockx=16, blocky=16, y0=16, y1=16, scthresh=12, micmatch=1, micout=False)
episode = core.vivtc.VDecimate(clip=episode, cycle=5, chroma=True, dupthresh=1.1, scthresh=15, blockx=16, blocky=16)

episode = core.std.SetFrameProps(clip=episode, _FieldBased=vs.FIELD_PROGRESSIVE)

episode.set_output()

I tried to use fine_dehalo, LGhost but without success.

Should I use some mask with another plugin ?

Best regards,

Klink
That isn't ghosting, these are halos that got resized.
Starting with the dvd source might be better,..
If you downscale and upscale the source you see that it doesn't really lose details, so that isn't the native resolution of the source; https://silentaperture.gitlab.io/mdbook-...aling.html might be interessting)

A few ideas:
  • downscale, then try some dehalo-filters (then upscale properly)
  • downscale and upscale with something like RealESRGAN or RealCUGAN
  • try aWarpSharp2 'luma only'
  • If you are on Windows, you could try some machine learning based model based filters. (through the vs-mlrt or torch-addon, later only if you have a newer NVIDIA card) Depending on the model, downscaling and upscaling through the model might be a good idea.
  • If you are on Windows with a fast NVIDIA card, you could try use BasicVSR++ (maybe masked), but that would be slow on 1080p content.
  • try to blur and then sharpen (or line darken)
Just to show the general idea, I uploaded a few examples to MediaFire.

Cu Selur
Thanks for your ideas and explanations.
I couldn't determine the native resolution on the tests I've made.
I will look further on it.
Also I will test on the bluray and dvd to see and thanks for the examples

Best regards,
Klink