Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Capturing and Editing Video > Avisynth Usage

Reply
 
Thread Tools Search this Thread Display Modes
Old 19th July 2013, 17:18   #1581  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
is Pel pertaining to whether your source is interlaced, telecined and/or progressive, or does it pertain to something else? I was under the assumption that Pel=1 was used if your source is progressive, and Pel=2 if it is interlaced.

But then in the FastDegrain doc, I read that the higher the Pel, the better the motion vector. Can someone please clarify this?

Thanks
unclescoob is offline   Reply With Quote
Old 19th July 2013, 17:21   #1582  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,346
Quote:
Originally Posted by unclescoob View Post
is Pel pertaining to whether your source is interlaced, telecined and/or progressive, or does it pertain to something else? I was under the assumption that Pel=1 was used if your source is progressive, and Pel=2 if it is interlaced.

But then in the FastDegrain doc, I read that the higher the Pel, the better the motion vector. Can someone please clarify this?

Thanks

No, it's motion estimation accuracy

Quote:
pel : it is the accuracy of the motion estimation. Value can only be 1, 2 or 4. 1 means a precision to the pixel. 2 means a precision to half a pixel, 4 means a precision to quarter a pixel, produced by spatial interpolation (more accurate but slower and not always better due to big level scale step). Default is 2 since v1.4.10.
poisondeathray is offline   Reply With Quote
Old 19th July 2013, 17:26   #1583  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Please consult http://avisynth.org.ru/mvtools/mvtools2.html

Pel is not related to progressive or interlaced material. It is used to tell how accurately the motion will be estimated. At pel=1, the super clip is not scaled at all, at pel=2 it is twice as big and at pel=4 it's four times as big. A bigger value for pel is not necessarily better as there's no way to scale an image by four times and not have any errors created by the extrapolation. Unless you live in the CSI world, that is Personally I would leave it at the default which is pel=2.

Basically I would leave all the values at their defaults as there's probably a reason why they are select. If you got the CPU power to spare, maybe use dct=5 and search=5 in the MAnalyse calls.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 19th July 2013, 19:52   #1584  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
I understand that using dct=5 uses the SATD (sum of absolute transformed differences) instead of SAD (sum of absolute differences) for luma. In your experiences, does this improve the motion estimation accuracy in a significant manner? for instance, in fast motion with animation, does it help in preventing blurring/ghosting at quick motion in dark scenes? I'm gonna try this for myself when I get home, but it's nice to hear from others while I'm stuck at this desk.
unclescoob is offline   Reply With Quote
Old 19th July 2013, 19:53   #1585  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
Thanks PDR and Boulder for explaining PEL
unclescoob is offline   Reply With Quote
Old 19th July 2013, 20:09   #1586  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
It's been quite a long time since I tested the differences, but if I recall correctly, the differences were mostly in the areas of higher contrast. I didn't test performance in the dark frames/areas much, but I think someone (was it *.mp4guy?) did some testing and found out that it helps there.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 19th July 2013, 20:12   #1587  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
Splendid!
unclescoob is offline   Reply With Quote
Old 19th July 2013, 20:12   #1588  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
Oh (as columbo would ask) one more thing, please folks if you have room for just one additional question: blocksize. 8 or 16? make any difference?
unclescoob is offline   Reply With Quote
Old 19th July 2013, 20:15   #1589  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Smaller is more accurate but more affected by noise. I personally use 8 with overlap 4 for all my stuff. If I use MRecalculate after that, I use blksize 4 and overlap 2 for SD stuff and blksize 8, overlap 4 for anything higher than that.
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 19th July 2013, 20:23   #1590  |  Link
unclescoob
Banned
 
Join Date: Jan 2013
Posts: 27
I use overlap 2. But in my script it's written as 8/2 (as per the way it's in the documentation). Is adding the / even necessary?

My stuff is analog SD and I'm using Mdegrain3. I'm thinking of using blcksize 4 if that's the case, since I'm prefiltering anyway.
unclescoob is offline   Reply With Quote
Old 19th July 2013, 20:54   #1591  |  Link
Boulder
Pig on the wing
 
Boulder's Avatar
 
Join Date: Mar 2002
Location: Finland
Posts: 5,718
Overlap 8/2 == 4
__________________
And if the band you're in starts playing different tunes
I'll see you on the dark side of the Moon...
Boulder is offline   Reply With Quote
Old 29th July 2016, 17:11   #1592  |  Link
real.finder
Registered User
 
Join Date: Jan 2012
Location: Mesopotamia
Posts: 2,587
what rfilter in MSuper do (Technical details, I know it did smoothing)? is it effect if pel = 1?
__________________
See My Avisynth Stuff

Last edited by real.finder; 29th July 2016 at 17:14.
real.finder is offline   Reply With Quote
Old 29th July 2016, 17:27   #1593  |  Link
feisty2
I'm Siri
 
feisty2's Avatar
 
Join Date: Oct 2012
Location: void
Posts: 2,633
Quote:
Originally Posted by real.finder View Post
what rfilter in MSuper do (Technical details, I know it did smoothing)? is it effect if pel = 1?
nothing
feisty2 is offline   Reply With Quote
Old 29th July 2016, 22:11   #1594  |  Link
kolak
Registered User
 
Join Date: Nov 2004
Location: Poland
Posts: 2,843
Quote:
Originally Posted by unclescoob View Post
I understand that using dct=5 uses the SATD (sum of absolute transformed differences) instead of SAD (sum of absolute differences) for luma. In your experiences, does this improve the motion estimation accuracy in a significant manner? for instance, in fast motion with animation, does it help in preventing blurring/ghosting at quick motion in dark scenes? I'm gonna try this for myself when I get home, but it's nice to hear from others while I'm stuck at this desk.
My finding which I posted in VS forum.

Search 6 and 7 works the best for me, based on the dominant movement in particular source/scene. I asked if this could be dynamically changed ahead of final calculations. 5 gives very different characteristic for final frames, never as good as when correct option 6 or 7 is chosen based on scene dominant movement.
Another interesting (or not) thing is fallback method. When you try interpolate low fps sources than quite often motion adaptive method fails giving huge artefacts. Some pro solution have fallback mode, which at some threshold switches to simple frame blending for given frames. In case of fps conversion (in current world of hybrid sources, e.g. 25p mixed with 50i) this is sometimes useful option.
Other than this mvtools results are very good. I'm thinking if anything more can be done. There is something called phase correlation, is this used in mvtools? I would like to see something reliably converting 24p to 25p, which is at the moment only a dream

DCT hits performance a lot, but sometimes helps on fades etc.

Last edited by kolak; 29th July 2016 at 22:13.
kolak is offline   Reply With Quote
Old 30th July 2016, 13:24   #1595  |  Link
real.finder
Registered User
 
Join Date: Jan 2012
Location: Mesopotamia
Posts: 2,587
Quote:
Originally Posted by feisty2 View Post
nothing
why it there then?

pel use resizer to upscale frame, is rfilter for the downsize to original dimension?
__________________
See My Avisynth Stuff
real.finder is offline   Reply With Quote
Old 30th July 2016, 13:36   #1596  |  Link
feisty2
I'm Siri
 
feisty2's Avatar
 
Join Date: Oct 2012
Location: void
Posts: 2,633
Quote:
Originally Posted by real.finder View Post
why it there then?

is rfilter for the downsize to original dimension?
yeah, rfilter stands for reducing filter
feisty2 is offline   Reply With Quote
Old 30th July 2016, 13:41   #1597  |  Link
real.finder
Registered User
 
Join Date: Jan 2012
Location: Mesopotamia
Posts: 2,587
Quote:
Originally Posted by feisty2 View Post
yeah, rfilter stands for reducing filter
and if pel = 1? did nothing?
__________________
See My Avisynth Stuff
real.finder is offline   Reply With Quote
Old 30th July 2016, 13:45   #1598  |  Link
feisty2
I'm Siri
 
feisty2's Avatar
 
Join Date: Oct 2012
Location: void
Posts: 2,633
Quote:
Originally Posted by real.finder View Post
and if pel = 1? did nothing?
pel = 1 means no upscaling nor downscaling (precision by a single pixel)
feisty2 is offline   Reply With Quote
Old 10th March 2023, 13:19   #1599  |  Link
DTL
Registered User
 
Join Date: Jul 2018
Posts: 1,041
Some not very fast but better answer - rfilter is used when generating smaller sized images for higher levels of search (levels=0 mean all possible levels).

It work at all pel (if levels count all or high enough). Disabling usage rfilter is for example pel=1 levels=1. So make only refine search at largest (full size) level.

The used filter may cause generating different MVs at each level (used as a predictor after interpolation at larger levels of search). So initial developers of mvtools put several different filters for reduction of size for smaller levels and also recommend to try filtering to smaller levels:
From documentation: You may also try to apply some external filter to superclip or its coarse bottom part (by appropriate crop and overlay).
DTL is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 20:54.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.