4:3 LB to 16:9 compression question

[ramzeva]

in the past along time ago whenever I used DVD-RB Pro to convert a 4:3 LB to true widescreen doing movie mode only it would use the entire disc, but lately it comes out the same size as it was before the conversion.. example

3.7gb Rocky 2 (movie only) comes out to 3.7gb after conversion, but in the past all movies I converted would use the entire disc... did something change or am I over looking some option?

just seems like it would benefit more using the whole disc considering its going to have more picture after cropping.

[wmansir]

It's probably due to this from v1.02:- Added "Dynamic Peak Analysis" which determines peak
bitrates from the original stream and uses the info
to set segment maximum bitrates. This should improve
quality by more accurately distributing available
bandwidth. It should also eliminate some possible
sources for playback stutter.

If your video is CBR(ish) than the max bitrate of the original will limit the size of your output.

You can try setting "max_bitrate=nnnn" in the [Options] section of the Rebuilder.ini file, located in the same folder as Rebuilder.exe, but I'm not sure if it will override the new feature. If not you can edit the Rebuilder.ECL file before starting the encoding step, change VBR_MAX_BITRATE= for each job listed.

Something like 7000 should be a safe value.

[jdobbs]

DVD-RB will not make the output larger than the source. That is important to ensure proper reintegration of the original audio streams.

It also generally isn't logical to try and make the output larger than the source... that would imply you can make a copy that is better than the original -- which, of course, is impossible. Increasing the bitrate would serve no purpose other than filling disc (without quality improvements). I could make it do that -- but I'd feel as if I were disengenuous in doing so.

[ramzeva]

I understand it wont improve the picture quality, just seems like when cropping off a lot of black to make it true widescreen could use more bitrate in the long run. Keeping the same bitrate but with more picture wont cause artifacts?

But yeah in the past the older versions would fill the whole disc.

[Boulder]

In this case it would be useful to allow as high bitrate as possible as the number of active film pixels is increased as ramzeva pointed out. Also, having more bits to spend would allow using a high bitrate quant matrix which is useful for minimizing detail loss during re-encoding.

[ramzeva]

thats what I was thinking, honestly it seems like you would lose quality from the lack of bitrate for the new pixels

[jdobbs]

Not really.

1. The sharp edges of the black border can actually increase the needed bitrate.

2. When you stretch the original to meet the new size, the newly created pixels are made of averages based upon the original. The "honest" resolution can never truly be higher than the original. Because of how they are created the new picture is highly compressible (basically at the same compression rate as the original).

[Boulder]

If the borders are mod-16 pixels wide/high, there'll be no sharp edges as complete macroblocks are being used. Even mod-8 borders are acceptable.

Have you done any testing regarding point 2? I've never done that myself but I find it hard to believe that a clip with a rather large increase in film pixels would still need about the same bitrate as the original one.

[jdobbs]

Try this:

1. Use CCE SP and set a Q for a picture that is 4:3 with 16:9 letterboxed inside. Then look at the quantization required to meet that Q.

2. Then do the resizing and use the same Q. I think you'll find little difference in the required quantization.

The underlying resolution never truly changes. It's like taking a gallon of water and putting it in a different shaped container -- it may look different but it's still a gallon. (An admittedly bad analogy -- but the best I could come up with)

[jdobbs]

Whoops. Forgot to respond to the first part of your answer. That would be true if the picture edge fell on boundaries that were 16 pixels high. But when a 16:9 picture is enveloped in a 4:3 frame -- they don't.

[Boulder]

Try this:

1. Use CCE SP and set a Q for a picture that is 4:3 with 16:9 letterboxed inside. Then look at the quantization required to meet that Q.

2. Then do the resizing and use the same Q. I think you'll find little difference in the required quantization.

The underlying resolution never truly changes. It's like taking a gallon of water and putting it in a different shaped container -- it may look different but it's still a gallon. (An admittedly bad analogy -- but the best I could come up with)
I'll definitely have to test this - as soon as I can think of a movie I have in my collection and is a letterboxed 4:3..I think 2010 is one of those crapolas.

[jdobbs]

You might want to try one that is truly 16:9. I think 2010 is 2.35:1.

I'll never understand what the authors were thinking when they take a classic film and hatchet it like that. It's an abomination.

[Boulder]

You might want to try one that is truly 16:9. I think 2010 is 2.35:1.
Yeah, you are probably right..looks like I'll have to visit the local library and see if I can come up with any test material. Many early Nordic releases were screwed up by letterboxing in 4:3 instead of a proper 16:9 release and the library seems to be full of the really old DVD stuff :)

2010 is a killer when it comes to compression. I think I squeezed it to something like 1500kbps avg bitrate and couldn't tell much difference to the original.

[scharfis_brain]

Some time ago, I recompressed the PAL release of 2010.
It was a DVD-5 but I needed to recompress it, cause it is 1:2,35 non-anamorphic with a field phase shift (looks ugly on PC).
I assume that it is a conversion from the NTSC-master.

So I recompressed it with fieldmatching, a dehalo filter and limitedsharpen. Looks pretty good now.
And like you said, it is pretty well compressable!

[laserfan]

Hmm, this thread has confused me (admittedly not hard to do at my age!).

I have done at least one LB-->16:9 anamorphic conversion which I'd set TargetSectors to DVD9 and it worked great. IIRC I used Lanczos4Resize. I burned the resultant upconversion to DVD+R DL. It certainly made sense to me at the time, particularly how the letterboxed source was much greater than DVD5, i.e. it seemed to me that Expanding the image then further Compressing it to DVD5 was adding injury to insult.

So ramzeva is right, something in the latest versions of RB Pro may not allow this? Even though his source was not >DVD5, it still seems that once you lop-off the black and resize to 720x480 you need more bits?

[jdobbs]

No. You're misinterpreting. This is nothing new -- its been there for quite a while (v1.02, October 2005).

The overall average bitrate is limited to no more than that of the original stream. If you are converting a DVD-9 to a DVD-5 that could typically never happen anyway.... it would be way to large. If you are converting a DVD-9 to a DVD-9 --> it would be pretty much exactly the same size as the original.

The discussion here is that someone wants to take a stream that is 3.7GB on the original, convert it, and make it 4.37GB on the output.

The point I'm making (or at least trying to make) is that this would be just throwing bits at the stream in order to fill the disc and give the the user a "warm fuzzy" -- but it really doesn't improve the picture. It also makes reintegrating the audio stream more risky. I could do that... but I don't think it would be honest in that they would think they were accomplishing something that they weren't.

Also -- I don't agree with your last statement. You aren't adding more bits. The size is 720x480/576 either way. The new size simply uses something other than black space where the borders are. But the new stuff is interpreted from the original -- which makes it more highly compressible (redundancy is a good thing in MPEG encoding).

[jdobbs]

By the way. If someone is really determined to do this... even against my advice -- the README file points out that you can override it with:

OVERRIDE_BITRATE_CHECK=1

But... don't complain when it causes problems... especially when you are doing it for what I consider to be no good reason. I seriously hesitated to even bring this up, because I can already see myself reading that line in someone's INI file who has sent me a nastygram about a "bug"...

[jdobbs]

One more point, now that we're way, way off topic...

The original question was about the undersizing... but at this point there is no real reason to imply that all this debate is even related to the problem presented.

What size is the original stream? Has anything else changed (encoder upgrade, changes in settings)?