Log in

View Full Version : 1080p v 720p, to DVD9.


liquidskin76
20th January 2010, 12:11
Hi All,

Just looking for a bit info/guidance/opinion regarding BD backups to MKV at 1080p/720p to DVD9 size.

If i do 2 backups of the same movie (movie only) to DVD9, using say BDRB or RipBot, at 720p and 1080p using the same quality settings, will the 720p backup be better quality based on higher bitrates v pixel increase of 1080p?

So, based on the above is it to be 720p to DVD9 or 1080p to DVD9?!! Would love to hear everyone's opinion!

Thanks

moviefan
20th January 2010, 12:23
Hi liquidskin76,

concerning BD movies' quality, I have noticed that 1080p resized to 720p (I do it with Spline64Resize) and sharpening afterwards (I do it with LSFMod) results in basically no loss in details while the picture looks equally sharp compared to the 1080p original. Dark Shikari reasoned this with e.g. lowpassing of the original BDs. (thread where this was discussed: http://forum.doom9.org/showthread.php?p=1365011#post1365011) However, I think it depends on the individual BD although I have not come across one that was worth keeping the 1080p resolution. If we're talking about movies of about 90 to 150 min length per movie, I would recommend to downsize to 720p as the extra bitrate per pixel keeps the usual grain a lot better and makes the picture look closer to the original (under the assumption that the downsize + sharpening does not take any actual detail).

liquidskin76
20th January 2010, 13:00
Hi moviefan,

Cheers for the info.

liquidskin76
20th January 2010, 13:21
Would love to hear Dark Shikari, Atak_Snajpera, jdobbs, and any other h264 guru's opinion!

moviefan
20th January 2010, 13:27
You're welcome! Oh, and maybe I should add: I usually have 2 audio tracks and, it depends on the movie, they sometimes are up to twice 1536kbps DTS which takes up a lot of bitrate only for audio (considering the size of a DVD9). It might be easier to keep the 1080p with only one audio track. You should try but I was shocked when I compared original 1080p to originalVideo.Spline64Resize(1280,720).LSFMod(preset="slow", kernel=19, preblur="on").SplineResize(1920,1080) and there was no difference. Sometimes the resized video was sharper (due to LSFMod, but in a natural way so it didn't look artificially sharpened) than the original without detail loss. Concerning the AVS chain: If there are letterboxes, I recommend: crop->downsize->sharpen->addborders for speed and quality.

liquidskin76
20th January 2010, 13:37
Are you able to post your avisynth config in correct format so i can copy? I've never used avisynth so would be cool to be able to shortcut!

Regarding audio, i usually just encode vid, then mux with original hd audio to mkv. The DVD9 size isn't really acurate however i'm playing from HDD so no real problem regarding disk space.

moviefan
20th January 2010, 13:54
Sure! My avisynth script is:


FFVideoSource("remuxed_source.mkv", cache=true, cachefile="remuxed_source.ffindex")

crop(0,132,-0,-132).Spline64Resize(1280,548)

src = last
replace = LSFMod(defaults="slow", preblur="ON", kernel=19)

ReplaceFramesSimple(src, replace, mappings="[0 XXX]") # XXX being end frame of movie (excluding credits) to speed things up when encoding the credits since they don't need to be sharpened

AddBorders(0,86,0,86)


You have to check the amount of pixels to crop for each movie individually and calculate the vertical resize respectively. Usually, it is just this setting or crop(0,140,-0,-140).Spline64Resize(1280,528) and later AddBorders(0,96,0,96).

If you're not targeting DVD9 very strictly, you should consider encoding in CRF mode which is only 1-pass (thus faster) and "adapts" to the complexity of your movie. Once you've found a CRF that suits you best, you might have movies that are smaller than DVD9, some are bigger.

liquidskin76
20th January 2010, 14:25
Thanks!

I do use CRF (usually ripbot CQ18 for bd). Spits out bd vid at about the 6 to 7GB mark, so final size of around 10GB with hd audio and mkv container overhead.

Do you use RipBot264? If so, do you do your DVD backups through it, and at what settings? I've been looking at CQ18 however not sure if that's overkill for DVD. Maybe CQ20?

Dark Shikari
20th January 2010, 18:19
I have yet to see a non-Pixar Blu-ray that actually had detail up to 1080p, so I would say "downscale to 720p unless there's a good reason not to".

moviefan
20th January 2010, 19:07
I only use x264 via command line meaning I created a batch file (Windows) so that I don't have to type in the settings every time. I do 2-pass encodes since I actually target DVD9 size encoding with Blu-ray compliant settings to make the encodes play as AVCHDs on a standalone Blu-ray player. However, if you only want to play them on your e.g. HTPC, you might want to consider different (less restrictive) settings that are suitable for you in terms of quality and speed. Maybe you want to ensure DXVA compliance which is less restrictive than Blu-ray compliance so you probably get better quality at same file size or the other way around same quality a smaller file size. The CRF value is from my point of view a parameter of personal taste. It depends how critical you are with the video quality and it depends a lot on your settings. I have experienced variations in the ratefactor (please correct me if I'm wrong that the ratefactor shown in the first pass of a 2-pass encode is more or less the CRF) by 6 at the same target bitrate only because of (very) different settings. But sticking to values around 18+-4 is reasonable in my eyes. You could cut out a sample of your movie (maybe 2-5 minutes or so) and encode it with different settings and CRF values to find out what you like.

Sharc
20th January 2010, 19:53
If one accepts 448kbps or 640kbps ac3 audio (good enough for my ears) one gets even on a BD-5 ratefactors in the range 17 ... 22 @ 720p with --preset medium for 100 minutes BD movies. At these ratefactors the the quality is very good.

Blue_MiSfit
20th January 2010, 20:14
@Dark_Shikari:

How do you determine how much detail a source actually has?

The only method I use is pretty simple:

a=last
spline36resize(1280,720).spline36resize(1920,1080)
interleave(a,last)

And then just step forward and backward in AvsP.

I'd agree that the vast majority of sources aren't worth 1080p. If anything, the grain softens a bit.. which irritates me, and leads me to believe a lot of it is artificially added. Nice touch ;)

Is there a better way to do this?

~MiSfit

MrCommunistGen
20th January 2010, 23:29
@Dark_Shikari:

How do you determine how much detail a source actually has?

The only method I use is pretty simple:

a=last
spline36resize(1280,720).spline36resize(1920,1080)
interleave(a,last)

And then just step forward and backward in AvsP.


Not sure if this is detailed elsewhere, but how do you actually go about doing this check? Do you just use an AviSynth script? I've never used AvsP but I'll check it out when I get a chance.

Thanks!

ImmortAlex
21st January 2010, 04:46
What about third option: resize to 1440x1080 and set SAR to 4:3? There is a bunch of movies, which fit into DVD-5 with one AC-3 track and at CRF ~22.

Blue_MiSfit
21st January 2010, 06:13
Not sure if this is detailed elsewhere, but how do you actually go about doing this check? Do you just use an AviSynth script? I've never used AvsP but I'll check it out when I get a chance.

Thanks!

I use AvsP to write this script, and then push F5 to render it. Then you can step forward and backwards with the arrow keys like in VirtualDub. It's just easier to have script editing and preview in the same app ;) AvsP is pimp stuff!!

~MiSfit

Raptus
21st January 2010, 11:13
How do you determine how much detail a source actually has?
If you transform the image into the frequency domain, via DCT or FFT for instance, the coefficients tell you if the content is actually using the resolution available.

rack04
21st January 2010, 15:51
Hi liquidskin76,

concerning BD movies' quality, I have noticed that 1080p resized to 720p (I do it with Spline64Resize) and sharpening afterwards (I do it with LSFMod) results in basically no loss in details while the picture looks equally sharp compared to the 1080p original. Dark Shikari reasoned this with e.g. lowpassing of the original BDs. (thread where this was discussed: http://forum.doom9.org/showthread.php?p=1365011#post1365011) However, I think it depends on the individual BD although I have not come across one that was worth keeping the 1080p resolution. If we're talking about movies of about 90 to 150 min length per movie, I would recommend to downsize to 720p as the extra bitrate per pixel keeps the usual grain a lot better and makes the picture look closer to the original (under the assumption that the downsize + sharpening does not take any actual detail).

Why do you sharpen when you resize?

lnatan25
21st January 2010, 17:33
I have yet to see a non-Pixar Blu-ray that actually had detail up to 1080p, so I would say "downscale to 720p unless there's a good reason not to".
Might I suggest The Dark Knight (IMAX scenes) and Transformers 2 IMAX Edition (it is amazingly dumb, but it looks good :)).

Brazil2
21st January 2010, 17:39
I have yet to see a non-Pixar Blu-ray that actually had detail up to 1080p
BARAKA (http://www.imdb.com/title/tt0103767/).

Biggiesized
21st January 2010, 19:35
I have yet to see a non-Pixar Blu-ray that actually had detail up to 1080p, so I would say "downscale to 720p unless there's a good reason not to".
The International (2009) - Columbia Pictures

or

Speed Racer (2008) - Warner Bros.

Sharc
21st January 2010, 20:01
What about third option: resize to 1440x1080 and set SAR to 4:3? There is a bunch of movies, which fit into DVD-5 with one AC-3 track and at CRF ~22.
Yes, I did it a few times. It's a Blu-Ray compliant compromise.

moviefan
21st January 2010, 20:15
Why do you sharpen when you resize?

Because the sharpeness of the picture after sharpening looks more pleasant to me. Without sharpening it's a little bit too smooth in my eyes.

Sharc
24th January 2010, 10:49
@Dark_Shikari:

How do you determine how much detail a source actually has?

The only method I use is pretty simple:

a=last
spline36resize(1280,720).spline36resize(1920,1080)
interleave(a,last)
.......
Is there a better way to do this?

*.mp4 guy posted his method and rationale here (http://forum.doom9.org/showthread.php?p=1296140#post1296140)
Works well with your proposed interleaving.

drob
14th February 2010, 17:27
I have yet to see a non-Pixar Blu-ray that actually had detail up to 1080p, so I would say "downscale to 720p unless there's a good reason not to".

What would be a good reason to keep the 1080 resolutions?

poisondeathray
14th February 2010, 17:34
What would be a good reason to keep the 1080 resolutions?

If your source has more than 720 lines of actual resolved vertical resolution (not frame dimensions)

Capsbackup
14th February 2010, 17:42
What would be a good reason to keep the 1080 resolutions?

IMHO, you should try/test both ways, watch the results of both on the screen you plan to use, and determine with your own eyes what looks better to you. :)
Personally, I like to keep the original resolution, and depending on the length, action, detail, etc... I will pick either BD-5 or BD-9 for the results.
I have found that it is rare to be disappointed with keeping 1080P, and most of my backups are BD-5, with one AC3 5.1 640kbps audio and one eng subtitle. X264 is that good! But everyone sees things differently. :p

drob
16th February 2010, 13:44
Would i be right in thinking then that a 1280x536@5 mbps encode would be better then 1980x800@10 mbps?

moviefan
16th February 2010, 14:22
"Better" is a too generalized word... Bitrate does not scale linearly with the amount of pixels. Most probably 1280x536@5Mbps would have similar quality to 1920x800@a little less than 10Mbps. However, it is still questionable whether keeping the full resolution is worth the additional space compared to a downscaled encode. If the video source has an actual resolution of 1920x800 it might be worth keeping the resolution. Most Blu-ray discs (if you are talking about them) don't have a real full HD resolution in my eyes. You can check this rather easily by comparing an original frame and a downscaled->upscaled one as the process of downscaling would take away details if there are any at given resolution.

drob
16th February 2010, 15:54
Your right about using "better", i was thinking it more in the direction of being on par with 1080. this seems to be the consensus on allot of discussions i have been reading.

rack04
16th February 2010, 15:57
Encode two files, one 1920x1080 and the other 1280x720. Keep the target bitrate the same. Build a simple menu with MultiAVCHD and switch back and forth and see if you can see the difference.

sumawo13
17th February 2010, 02:07
If you transform the image into the frequency domain, via DCT or FFT for instance, the coefficients tell you if the content is actually using the resolution available.

Is there a program that does this?

Blue_MiSfit
17th February 2010, 05:26
Indeed! I wonder the same thing.

~MiSfit

bizz & buzz
19th February 2010, 17:32
If you transform the image into the frequency domain, via DCT or FFT for instance, the coefficients tell you if the content is actually using the resolution available.
isn't that the way to find the minimum possible resolution without loosing any real details, with mathematical precision?
i.e, encoders heaven?

please guys, do elaborate on this.

julius666
19th February 2010, 21:55
isn't that the way to find the minimum possible resolution without loosing any real details, with mathematical precision?
i.e, encoders heaven?

please guys, do elaborate on this.

Well, I'm not a video expert, but I'm afraid that compression (e.g. blocking noise) and dithering screws up the frequency domain.

Raptus
22nd February 2010, 20:30
Well, I'm not a video expert, but I'm afraid that compression (e.g. blocking noise) and dithering screws up the frequency domain.
If you choose a DCT block size that matches the sources smallest block size, you avoid this. Dithering is tricky, but it's low intensity (low coefficient values), basically creating a noise floor, which can be dealt with. Also this only affects lossy sources.

The analysis could be done with 2D histograms (one axis for coeff. index, the other for coeff. value, bar height proportional to number of occurrences; visualization in 3D), one for each image axis. I don't have Matlab at my disposal atm to try it out...

Blue_MiSfit
22nd February 2010, 21:11
I figured matlab would be a good tool for these kind of analysis. Sadly, I know nothing of it :)