View Full Version : If you have downsized 1080 to 720, what scaler and filters did you use?
thrakk
20th September 2011, 04:00
For those of you that have downsized 1080 to 720, what scaler and filters did you use to clean your finished video?
Can you share some of your code with me?
I'm converting a Blu-ray to a DVD. File size isn't much of an issue. I'm going for clarity here.
(and I'm not using animated video footage)
aegisofrime
20th September 2011, 04:53
I would like to know as well. I have some 1080i footage that don't look so good at that resolution, so I'm thinking of converting to 720p. I'm wondering if a simple Spline36Resize(1280,720) after deinterlacing would suffice.
Blue_MiSfit
20th September 2011, 07:56
spline36 is more than enough.
I often use bicubic for speed reasons.
Derek
hello_hello
20th September 2011, 08:55
I'm generally a "no filter" kind of guy. There'd have to be a fairly specific reason for me to try to "improve" on the source rather than just duplicate it.
Either Lanczros or Spline36 resizing (although to be honest I can't really see much of a difference) and some very minor noise removal (undot.dll) would be about it for me.
If file size isn't much of an issue just use CRF encoding with a CRF value of somewhere around 18 to 20. Or if you're wanting to burn the encode to a DVD... use all of it.
madhatter300871
20th September 2011, 10:42
I use Spline36, although I cant really see a difference between it and Lanczos4 either. Time is not a premium for me, I just set it going and leave it. For me a 2 hour film takes about 3 hours to re-encode (approx).
I don't use any filters to clean up a blu ray source.
If size isn't an issue, you would not want to re-encode. So presuming you want to fit your final encode onto a single sided DVD, use 2 pass encoding. I use 2 pass and can fit a 2 hour movie onto a single sided DVD in full HD, quality is very very good. I do, however, only use the core DTS track or 5.1 AC3 and never the DTS master audio.
A 1280x720 re-encode will fit onto a single sided DVD no problem at all, and I would more than likely do a CRF encode with a value of 18, perhaps 20 or 22 for a particularly long movie.
Anything more than 2 hours-ish and I use double sided DVD's.
Emulgator
20th September 2011, 12:23
An almost perfect solution I can recommend and use (especially when it comes to interlaced resizing):
hd2sd() for down, sd2hd() for up.
The placebo interlaced downsizer solution:
(Q)TGMC()
#then use the resizer you prefer, reinterlace.
separatefields.selectevery(4,0,3).weave# This is for interlaced output, for bobbed output comment this line out.
My favourite placebo upsizer (I only use this one now, just beautiful):
<your sourcefilter here>("X:\.avi")
ConvertToYV12(interlaced=true)
TempGaussMC_beta1mod(EDIMode="NNEDI2") #or nnedi3 (I modded my TGMCs to accept nnedi3)
nnedi(field=0,dh=true).TurnRight().nnedi(field=0,dh=true).TurnLeft()#This doubles horizontal/vertical size, or use _rpow2
#BlindDeHalo3(rx=3.2,ry=1.5)
#sharpen(0.3)
spline36resize(1440,1080)# or what one may need (1920,1080) (1280,720) etc.
separatefields.selectevery(4,0,3).weave# This is for interlaced output, for bobbed output comment this line out.
Yellow_
20th September 2011, 21:39
I'm generally a "no filter" kind of guy. There'd have to be a fairly specific reason for me to try to "improve" on the source rather than just duplicate it.
This is something I've been meaning to ask. So to take the opportunity, I've seen applications that do point resize or fast gaussian to go from YCC to RGB / composite and color correct in RGB and then encode out to YCC.
Would that be a reason to use something more than just duplicate going to RGB or would any edge refinement be done in RGB before going back to YCC?
@Emulgator are those hd2sd() & sd2hd(), the work of Dan Issacs over at the Adobe forums?
nibus
20th September 2011, 22:24
Most filters will just make a bluray source look worse... Plus your encode will take much much longer with extra filters.
But undot() is great to increase compressibility without any noticeable degradation or speed decrease.
2Bdecided
21st September 2011, 00:09
The placebo interlaced downsizer solution:
(Q)TGMC()If you're downsizing interlaced HD to interlaced SD, the choice of deinterlacer is completley irrelevent - bob() vs qtgmc() - the difference is invisible after re-sizing and re-interlacing.
Cheers,
David.
Emulgator
21st September 2011, 03:00
@Emulgator are those hd2sd() & sd2hd(), the work of Dan Issacs over at the Adobe forums?
Yes, it is Dan Isaacs.
thrakk
22nd September 2011, 21:13
well I was wrong about 1080...
I'm going from 720p to 480p and both input and output are YUV 4.2.0
I think undot()/removegrain(mode=1) might be helpful. I don't know if it would be too much smoothing.
Should I use a filter to avoid chroma artifacts?
Is there any additional things I can do about hue-equalization?(besides the ColorMatrix switch and any manual tweaking of brightness, contrast and saturation levels)
hello_hello
22nd September 2011, 22:30
When I tested SeeSaw (http://forum.doom9.org/showthread.php?t=104701&highlight=SeeSaw_2006.01.02) a while back I was fairly impressed. How much effect it has seems to depend on the source quality, although I only ran a few encodes with it so I may have no idea what I'm talking about. One of my test 720p to 480p encodes using a fairly average quality source, turned out surprisingly well. In some areas the 480p encode looked better than the source.
I don't actually use SeeSaw (http://forum.doom9.org/showthread.php?t=104701&highlight=SeeSaw_2006.01.02). It's nice, but it slows the encoding process down quite a lot.
thrakk
23rd September 2011, 19:12
Thanks for the replies everyone. I'm sure that your answers will help others wanting to perfect their downscaling filter chain
Motenai Yoda
24th September 2011, 02:44
I use blackman taps 6
and in case fft3dgpu with slight settings or bsharp
SilaSurfer
27th September 2011, 17:10
HighPassSharp(r=0.15) # My Hi-pass sharpen implementation with a Blur=0.15
@Rean
Could you post this function here? Thanks
henryho_hk
28th September 2011, 04:06
Is it interlaced-to-interlaced, interlaced-to-progressive or progressive-to-progressive resize?
SilaSurfer
29th September 2011, 16:16
Thanks Rean. I'm going to try it out. Do you chain those three versions one after another, I'm asking because HighPassSharp has one radius setting the other one has two etc?
Undead Sega
5th October 2011, 02:45
I use:
Blackmanresize
I dont know if this is a recommended thing to continously use but from what I know, it's the best I've used so far.
*.mp4 guy
5th October 2011, 07:00
I switched to using blackmanresize(taps=4) for downsizing most content a while ago. Spline36 is more bulletproof for avoiding aliasing, but blackman taps=4 can look really nice if aliasing isn't an issue.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.