Vapoursynth DotKill

From wiki.selur.net
Jump to navigation Jump to search

DotKill removes dot crawl and, in its telecine-aware modes, rainbows from composite-sourced video. It has a spatial mode for general use and two modes for NTSC 3:2 pulldown material such as anime.

Filtering → Vapoursynth → DeCrawl

DotKill controls in Hybrid
DotKill controls in Hybrid

What it does

  • spatial (DotKillS) cleans the luma within each frame; it can be used on most material after field matching.
  • pseudo-spatial (DotKillZ) and spatio-temporal (DotKillT) are meant for 3:2 pulldown material whose rainbows were added after the pulldown. They work on the telecined fields, so they belong before field matching/IVTC. The spatio-temporal mode blends with the previous or next field where blocks did not change.

Settings

Control Parameter Range / values Hybrid default Meaning
Mode – spatial, pseudo-spatial, spatio-temporal spatial Selects DotKillS, DotKillZ or DotKillT.
Iterations iterations 1 – 10 1 How often the spatial filter runs; 1–4 is usually best, more can cause artifacts. Also used when a temporal mode falls back to spatial on progressive input.
Order order 0 – 1 0 Field order (pseudo-spatial and spatio-temporal); 1 is untested upstream.
Offset offset 0 – 4 0 Position in the pulldown cycle (pseudo-spatial and spatio-temporal); found by trial and error.
DupThresh dupthresh 0 – 256 64 Spatio-temporal: difference above which a block counts as changed between fields; 32–128 is usual, 0 behaves like pseudo-spatial. Too high values can cause heavy artifacts.
TRatio tratio 1 – 1024 3 Spatio-temporal: if more than 1/TRatio of the blocks changed, temporal filtering in that direction is skipped; higher means fewer artifacts in motion.
Show show on/off off Spatio-temporal: marks the blended blocks (white = next frame, black = previous).

Example

Default settings on a 10-bit YUV 4:2:0 source (640x356):

clip = core.dotkill.DotKillS(clip)

How Hybrid calls it

  • Input is YUV 4:2:0 at 8–16 bit integer; the spatial mode also accepts gray. Other formats are converted first.
  • If a temporal mode is selected but the input is progressive, Hybrid uses the spatial mode with the Iterations value and notes this in a script comment.
  • The temporal modes need the field-based material, so DotKill has to run before deinterlacing/field matching in the chain.
  • DupThresh is entered on the 8-bit scale; the plugin scales it.
  • Show is only written in preview scripts, never in the encode.
  • Values are only written when they differ from the defaults, which equal the plugin's own defaults.
  • DotKill also offers the common options (Masked, ApplyOnlyTo, Merge, OnRegion, Gimmick).

Background

Plugin details

DotKill is by Fredrik Mellbin (myrsloik); release R4 added 16-bit support. The spatial mode processes only luma and clamps it to the limited range; results above 8 bit are not bit-identical to 8-bit results. The pseudo-spatial mode only cleans every other output frame. Offset 4 in the spatio-temporal mode filters both fields separately. The frame property DotKillTOffset helps to check the chosen offset.

Typical workflow

The upstream suggestion for telecined material: DotKillT before VFM, DotKillS after VFM, then VDecimate.

Sources