I got to thinking about various codec issues being discussed in these forums. We all know various complaints that people have with different video codecs and their deficiencies. MPEG-2 has horrid macroblocking with motion when their isn't enough bits thrown at it. We've all seen it on OTA and on some early Blu-ray discs.

Some complain that VC-1 can look very soft like the images may have been pre-filtered taking out the grain.

My personal observation with both VC-1 and AVC is that both can look very sharp with realistic looking film grain, but there have been soft transfers with both codecs.

Amir and Ben on the insider's thread have stated that soft transfer with VC-1 is not due to any codec deficiency, and that VC-1 can be just as sharp as any AVC encode. The problem according to them (forgive me if this is oversimplifying or simply wrong) is that although AVC does very well with static or slow moving images, that it softens up when there is a lot of motion. Similar to how MPEG-2 macroblocks with motion, AVC softens the image to compensate for lack of bits. Perhaps is why there is lot of conflicting report on PQ of various AVC encodes. When you see close-up facial features or slow pans, you see lots of details, but perhaps the picture does deteriorate much more than VC-1 when there is a lot of motion.

How would you test if this is true? I know I can pick out macroblocking, but I am not sure that I would be able to pick up softening of image when there is lot of motion or explosion going on.

I got to thinking about various codec issues being discussed in these forums. We all know various complaints that people have with different video codecs and their deficiencies. MPEG-2 has horrid macroblocking with motion when their isn't enough bits thrown at it. We've all seen it on OTA and on some early Blu-ray discs.

Some complain that VC-1 can look very soft like the images may have been pre-filtered taking out the grain.

My personal observation with both VC-1 and AVC is that both can look very sharp with realistic looking film grain, but there have been soft transfers with both codecs.

Amir and Ben on the insider's thread have stated that soft transfer with VC-1 is not due to any codec deficiency, and that VC-1 can be just as sharp as any AVC encode. The problem according to them (forgive me if this is oversimplifying or simply wrong) is that although AVC does very well with static or slow moving images, that it softens up when there is a lot of motion. Similar to how MPEG-2 macroblocks with motion, AVC softens the image to compensate for lack of bits. Perhaps is why there is lot of conflicting report on PQ of various AVC encodes. When you see close-up facial features or slow pans, you see lots of details, but perhaps the picture does deteriorate much more than VC-1 when there is a lot of motion.

How would you test if this is true? I know I can pick out macroblocking, but I am not sure that I would be able to pick up softening of image when there is lot of motion or explosion going on.

You'd only be able to tell if you have something to compare against - either the original or a different encode. Example MI:3 showed something interesting between mpeg2 & vc-1 that could be due to compressionist error or something in the codec that made it easy to generate problem segments like that.

You can encode some 1 hour HD programming in AVC at 3.5Mbps quite easily (if it was a talking heads interview) and the user won't know that some very bad things are going on underneath because of the heavy deblocking going on, but if you have the original, it would be night and day.

Often, the best places to suspect something is wrong is to pay a lot of attention to leafs on trees or faces or edges of clouds, in low bitrate encodes, with all the advanced codecs' ability to hide the encoding artifacts, often the behavior around pores of people's faces is the best way to notice something is obviously wrong, and even then you can't be always sure if it's just too much makeup or out of focus shots.

Frame by frame though, you should be able to tell if the bitrate falls below a certain threshold, since that's when things get quantized too much and that's when the advanced codecs' overlap smoothing and deblocking filter start to earn their pay by hiding all the bad stuff around block boundaries.

If you take away all the rhetoric, the facts are that these in loop filters work like smoothing erasers, or smudge tools around edges to hide problems in the encode. They cannot get closer to the original image this way, only hide what was done in the encode and present an aternate frame of what should look 'natural' even though it no longer is the same as the original.

This is an interesting element, and I suspected as much that something like this was going on with vc-1/avc/mpeg-4 family of codecs. It's not so much that these advanced codecs are "immune" to breaking up under bitrate starvation. They just have an extra behavior that attempts to cover up the fact, once they are breaking up.

What I also wonder is if there would be any benefits, if instead of deblocking macroblocks, a dynamic blurring filter could be applied ahead of the encoding to ease the hit on bitrate starvation? That is, once the encoder reaches some macroblocking threshold on a particular frame due to bitrate starvation, the original frame is resubmitted to a blurring filter and reencoding (and tested again for a macroblocking threshold) and then proceeding with successive frames similarly until bitrate has "caught up" again. I don't know if one way or the other would be better (or better in different ways), or would the result be pretty much the same in pq. "Pre-blurring" may seem like a bad idea, but if the scene is going to be in high motion (which would cause a break-up, anyway), a good amount of blurring should be tolerable w/o adversely affecting the performance as a whole.

The theme here is, is there any advantage to doctoring the frame before an artifact has a chance of occuring vs. attempting to hide/obscure the artifact after it occurs?

A dynamic prefilter would not be adaptive enough to work properly. Note that the smudging around edges adapts to the amount of blocking going on. If bitrates are high enough, no smudging occurs.

High enough depends on the stream, as some streams are torture segments while others are very easy.

Why couldn't it be "adaptive" enough?

Besides motion video sequences, such as those being proposed for the 'codec-shootout' disc in another thread, perhaps capturing rotating/static resolution-wedge test patterns would add useful information. That's what was used in the mid-'90s by video experts okaying the ATSC (see their final report (http://www.atsc.org/news_information/papers/1995_acats/tsreport.pdf)). Resolution wedges at various rotation speeds would provide hard numbers for both luma and color (see table 2.3 in the report). They used speeds up to 5 rpm. Data obtained might correlate with observations of ordinary motion-video sequences. -- John

I would like to suggest a test involving flashing lights (or lighting effects that have a rising/falling effect on objects), as well.

Also, a test that allows the assessment of the smoothness of a large gradient.

Why couldn't it be "adaptive" enough?

The lossyness of an encode is really only dependent on the quantization, actually around the boundaries of each transform/predicted mode block. When quantization is low, hardly anything is lost, and you don't want to turn on the edge smudging filters --- this actually happens a lot (not having to turn on the smudging).

A prefilter would not know how each macroblock is quantized, and not even how each macroblock is going to be subdivided, as such, it will tend to flter out details that would have been encoded very well if allowed to go through.

That is not to say that someone could be smart enough to invent something that allows you to perhaps in a statistics gathering phase, do a preencode with the dynamic filter and make everything work.

Based on what I see though, around 40Mbps it is possible to preserve a lot of detail even with tons of grain, and it would be a mistake to prefilter them just to fit to a lower peak envelope.

---

Besides flashing lights, (btw, there is a very good tool in h.264 to deal with flashes of cameras, but it is almost impossible to implement well without a human playing with a few parameters), my favourite is the water sprinkler - add some slight movement in a human face in the foreground and you have a really good torture test.

All this talk about VC1 being "softer" than AVC is just not true - in fact it is the reverse, based on what I see today. Top AVC encodes I've seen are usually softer than top VC1 encodes, although there is visibly more "noise" or "grain" in the AVC encodes. Mpeg2 on the other hand is either soft or noisy, I haven't seen much in between...

I am going to add links to some images for you to make your own judgement...

--- Ready!

I'm linking to some images that Xylon has kindly and painstakingly captured digitally directly from the encodes and posted in his thread here: http://www.avsforum.com/avs-vb/showthread.php?t=811102&page=1&pp=30

Here are sample shots from VC1 (Harry Potter and King Kong), MPEG2 (Underworld Evolution) and AVC (XMen 3).

I believe that the Mpeg2 shots looks much softer than the VC1 shots, and the AVC shots look much more grainy, yet still not as sharp as VC1.

NOTE! - Turn off "Automatic Resizing" in your Internet Explorer (or other) browser, or you'll see a downsized version of these images which will not let you see the detail. (Go to Tools/Internet Options/Advanced to make this change)...

NOTE! - Make sure you use the bottom scroll bar in your browser so you can see the whole image all the way out to the right...

Let your own eyes be the judge....


Sample VC1 Shot from King Kong:

http://www.imageviper.com/displayimage/80635/0/KKhddvd1.jpg


Sample VC1 Shot from Harry Potter:

http://i166.photobucket.com/albums/u98/adzez/HD%20DVD/8186d51e.png


Sample MPEG2 Shot from Underworld Evolution:

http://i166.photobucket.com/albums/u98/adzez/Bluray/e6122f50.png


Another Sample Mpeg2 Shot from Underworld Evolution:

http://i166.photobucket.com/albums/u98/adzez/Bluray/29264984.png


Sample AVC Shot from XMen 3:

http://uploadingit.com/files/8540/Bluray/XmenIII8.png


Another Sample AVC shot from XMen 3:

http://i166.photobucket.com/albums/u98/adzez/Bluray/cdb0d4a3.png


Just for comparison, another VC1 Shot from King Kong: Look at how crisp and sharp the image is, and how little compression noise there is...

http://i166.photobucket.com/albums/u98/adzez/HD%20DVD/f13dd237.png

A prefilter would not know how each macroblock is quantized, and not even how each macroblock is going to be subdivided, as such, it will tend to flter out details that would have been encoded very well if allowed to go through.

That is not to say that someone could be smart enough to invent something that allows you to perhaps in a statistics gathering phase, do a preencode with the dynamic filter and make everything work.
I think there are some MPEG-2 encoders that do stuff along these lines already, particular Canopus ProCoder. At least that's what it looks like what's happening in the output to me.

Basically a "pre-baked" in-loop deblocking filter. Not as good as the real thing, but still something.

Besides flashing lights, (btw, there is a very good tool in h.264 to deal with flashes of cameras, but it is almost impossible to implement well without a human playing with a few parameters), my favourite is the water sprinkler - add some slight movement in a human face in the foreground and you have a really good torture test.
We've got a great thing in our VC-1 implementation for flashing lights...

Sprinklers are really painful. My nightmare is foliage blowing in the wind (sharp edges, random motion, green is the major Y' component). I about tore my hair out years ago encoding some Robert Redford interview footage from Legand of Bagger Vance, where he was interviewed in front of a hedge on a windy day. I had to use literally triple the data rate for that sequence compared to others from the same tape.

There - I've added some of these shots to my post above. Samples from Mpeg2, AVC and VC1 titles...

You be the judge...

rdjam,

The first shot from Underworld is actually from Harry Potter.

A prefilter would not know how each macroblock is quantized, and not even how each macroblock is going to be subdivided, as such, it will tend to flter out details that would have been encoded very well if allowed to go through.

This is why I specified that the action of the filter would be governed by whatever macroblock stats you collect after encoding the frame via normal settings. If the stats are such that the filter should be engaged, then the original frame is "preblurred" by some amount and then resubmitted to the encoder. The loop is repeated until just the right amount of preblur is present to get an encoded frame that behaves nicely with the available bitrate.

It just strikes me that the solution that averts the problem before it can occur should usually return higher quality results than having the problem occur and then cleaning it up after the fact.

rdjam:

Look I got destroyed for this when I brought up Lady in the Water, and only CITED other films as examples for illustration. But here's the problem...

(1) These ain't the same movie.
(2) They are sinlge frames... ie, could I find a sharper frame in UW:E and a softer one in KK? You bet.
(3) They're screenies, and often do not represent overall quality of a moving picture.

I would say these things no matter whose codec you were trying to prove better. Also, a 4th element that PaidGeek mentioned (I think) is that the compressionist influences the encode too. So even though you compare codecs, there's still a human element.

Benwagonner, I would hope you'd chime in here and confirm some of these issues, as discussed in the insiders thread.

-Pie

Not only that but everytime someone makes a comparison to show a flaw in VC1 they get the standard reply that you cannot compare the PQ between two different films. (which I happen to agree with). That same logic applies here too.

Try again rdjam.
"Try again"?...

Try not to be so dismissive.

You will rarely see the same movie in different codecs on BD and HD.

However, regardless of the movie, you can clearly see the difference in what VC1 is capable of, in comparison to the other two formats.

For anyone to try to discount this difference in quality as not relevant simply because you don't have the same movies available in each is just an attempt to never let people see for themselves.

If anyone has ANY mpeg2 or AVC content from either HD DVD or Bluray that is better than the VC1 shots you are seeing above, then let me know what it is. I'll be happy to check it out.

VC1 is the benchmark for what is on sale for either format.

rdjam,

The first shot from Underworld is actually from Harry Potter.
Darned cut and paste gets confusing with all these links! :) Now fixed....

Thanks!

rdjam:

Look I got destroyed for this when I brought up Lady in the Water, and only CITED other films as examples for illustration. But here's the problem...

(1) These ain't the same movie.
(2) They are sinlge frames... ie, could I find a sharper frame in UW:E and a softer one in KK? You bet.
(3) They're screenies, and often do not represent overall quality of a moving picture.

While I do not disagree with the theory, I would state that these factors do not prevent one from assessing the "capabilities" and "limitations" of each of the codecs, through visual assessment.

Again, if Mpeg2 and AVC can match the VC1 images above, there should be movies that someone can cite "right now" that we can capture some sample images from, which could demonstrate the prowess of the other codecs.

Right from the release of Phantom of The Opera at the very first days of HD DVD, VC1 has demonstrated a capability that I have not yet seen in any of the Mpeg2 or AVC movies I've watched to date.

Oddly enough, the best AVC encodes I have seen to date were from SkyHD in the UK... (but the efficiency seemed to be gained by liberal "clipping" of brighter areas.)

Amir and Ben on the insider's thread have stated that soft transfer with VC-1 is not due to any codec deficiency, and that VC-1 can be just as sharp as any AVC encode. The problem according to them (forgive me if this is oversimplifying or simply wrong) is that although AVC does very well with static or slow moving images, that it softens up when there is a lot of motion. Similar to how MPEG-2 macroblocks with motion, AVC softens the image to compensate for lack of bits. Perhaps is why there is lot of conflicting report on PQ of various AVC encodes. When you see close-up facial features or slow pans, you see lots of details, but perhaps the picture does deteriorate much more than VC-1 when there is a lot of motion.
Well, it isn't that much to do with motion per se. It's just that AVC will get softer under bandwidth stress, at rates where VC-1 can maintain detail.

How would you test if this is true? I know I can pick out macroblocking, but I am not sure that I would be able to pick up softening of image when there is lot of motion or explosion going on.
This would be hard to do without access to the original source to compare.

Can we get back to the topic. Let's assume that AVC's deficiency is softening under lots of motion. How would you best demonstrate such artifacts? I think with any codecs, you can achieve a very sharp image when things aren't moving. It's when there is a lot of motion, that you really test how good the compression algorythm works.

I can post a ton of soft screenshots from The Departed and a ton of sharp screenshots from Crank.Then what are you waiting for? ;) By the way, doesn't that statement alone indicate a recognition that The Departed is a much sharper picture overall than Crank? :)

What will that prove? Nothing. Actually it would prove that AVC and Mpeg2 can match VC1 quality - if the shots actually do compare.

And what does your comparison prove? Nothing.At the moment it seems to show a pretty clear VC1 lead in picture quality. But again, feel free to name your challenger, in the form of an HD/BD movie in mpeg2 or AVC that matches the VC1 shots above...

When I saw the trailer for "King Kong" on the Toshiba demo disk that was reportedly done with AVC I had a good idea that it would be one of the best looking things out there, regardless of which codec ended up getting used for it (especially VC-1 or AVC/MPEG-4). It might be interesting if we could get ahold of that one (although it would have been done with their encoder from at least about a year ago).

--Darin

Can we get back to the topic. Let's assume that AVC's deficiency is softening under lots of motion. How would you best demonstrate such artifacts? I think with any codecs, you can achieve a very sharp image when things aren't moving. It's when there is a lot of motion, that you really test how good the compression algorythm works.
I fear we may never get our hands on the source material for the movies now available in AVC and the other formats, for comparison.

As such, the best option may be to find equivalent fast motion scenes from various assorted movies and capture multiple frames for comparison.

This will probably help us identify how their "failure modes" manifest themselves.

I think most of us know how mpeg2 fails under these circumstances, since mpeg macroblocking is rather infamous, but it'll be interesting to see the other two formats' results.

The MPEG2 version of MI:3 outdoes the VC1 version (see here (http://www.avsforum.com/avs-vb/showthread.php?t=813105)), which specifically shows macroblocking. There are a large number of releases--including almost every Paramount release--that are MPEG2 on one format and VC1 on another, despite your statement that they are "rare":

• Good Night, and Good Luck
• Kiss Kiss Bang Bang
• Training Day
• Full Metal Jacket
• Sleepy Hollow
• Aeon Flux
• Sahara
• Reds
• The Italian Job
• Nacho Libre
• World Trade Center
• We Were Soldiers
• Manchurian Candidate

I'm sure I'm missing a couple, too. I don't see why you can't post direct comparisons. Unless they're trying to sell you something, almost anyone will tell you to compare apples and apples. You're a consumer, so you're not trying to sell us anything.

Right?

We should try to come up with a plan to implement such rigorous testing on that ultimate test disc that Richard from RB Films is planning. Perhaps that foliage blowing in the wind that Ben mentions above. I know that particular shot on Fearless HD DVD disc looked very soft and suspected something was amiss. If we can't get commercial footage, how expensive can it get to get couple minutes of volunteer shot footage of foliage blowing in the wind?

I believe that "Babel" is being released by StudioCanal soon and they usually use VC-1. If so, maybe we can check whether Ben's claim about AVC/MPEG-4 not being able to do film grain will hold up against VC-1.

--Darin

I believe that "Babel" is being released by StudioCanal soon and they usually use VC-1. If so, maybe we can check whether Ben's claim about AVC/MPEG-4 not being able to do film grain will hold up against VC-1.

--Darin
Again, film grain issue may only manifest during fast motion scenes. I've seen plenty of film grain with all three codecs.

The MPEG2 version of MI:3 outdoes the VC1 version (see here (http://www.avsforum.com/avs-vb/showthread.php?t=813105)), which specifically shows macroblocking. But no proof that what is shown in that thread isn't a "decoding" issue rather than an "encoding" issue.

By the way, I'm curious as to how those screen shots were captured also. It'll be interesting to see if Xylon's digital stream captures will show the same effect, since they show more closely what has actually been encoded...

There are a large number of releases--including almost every Paramount release--that are MPEG2 on one format and VC1 on another, despite your statement that they are "rare":As you state, I said "rare", not non-existent... ;)

And, yes, these films will make for some very interesting comparisons over the next few weeks...

I don't see why you can't post direct comparisons. I promise I'll do just that ;) looking forward to it, actually...

Again, film grain issue may only manifest during fast motion scenes. I've seen plenty of film grain with all three codecs.
I usually notice the opposite. Fast motion creates softness, whereas scenes with limited motion have demonstrated the most grain noise.

Film grain is certainly natural, but some encodes have a propensity to make the natural grain look pretty awful, whereas others look very smooth and somewhat imperceptible (natural).

I usually notice the opposite. Fast motion creates softness, whereas scenes with limited motion have demonstrated the most grain noise.

Film grain is certainly natural, but some encodes have a propensity to make the natural grain look pretty awful, whereas others look very smooth and somewhat imperceptible (natural).
That's what I meant. Film grain issue being lack of film grain.

Maybe that's how we could test this issue. See how much film grain disappears during fast motion sequences. Film grain is inherent in the film, so it shouldn't have anything to do with focus or motion (would completely out of focus shot of moving white wall still have film grain?), correct? If we see baseline film grain during part of the movie with little motion, but if that decreases during fast motion, can we attribute that to bit starvation? This would only work if the picture softens globally rather than just parts of the picture with motion. Anyone know?

All this talk about VC1 being "softer" than AVC is just not true - in fact it is the reverse, based on what I see today. Top AVC encodes I've seen are usually softer than top VC1 encodes, although there is visibly more "noise" or "grain" in the AVC encodes. Mpeg2 on the other hand is either soft or noisy, I haven't seen much in between...

I am going to add links to some images for you to make your own judgement...

--- Ready!

I'm linking to some images that Xylon has kindly and painstakingly captured digitally directly from the encodes and posted in his thread here: http://www.avsforum.com/avs-vb/showthread.php?t=811102&page=1&pp=30

Here are sample shots from VC1 (Harry Potter and King Kong), MPEG2 (Underworld Evolution) and AVC (XMen 3).

I believe that the Mpeg2 shots looks much softer than the VC1 shots, and the AVC shots look much more grainy, yet still not as sharp as VC1.

NOTE! - Turn off "Automatic Resizing" in your Internet Explorer (or other) browser, or you'll see a downsized version of these images which will not let you see the detail. (Go to Tools/Internet Options/Advanced to make this change)...

NOTE! - Make sure you use the bottom scroll bar in your browser so you can see the whole image all the way out to the right...

Let your own eyes be the judge....




:D Now I know why it is called AV SCIENCE forum :D

:D Now I know why it is called AV SCIENCE forum :D
Eyes aren't used in science too? ;)

The entire science of A/V is to bring a more pleasing presentation to the end user. So it's no surprise that if someone finds an image more pleasing, they're inclined to prefer that one, no?

But no proof that what is shown in that thread isn't a "decoding" issue rather than an "encoding" issue.
Wouldn't that then also apply to your examples?

Wouldn't that then also apply to your examples?
No - the error being shown in the other thread was clearly an encoder/decoder issue, and was visibly not was intended to be there. Two different decoders were used to make those comparisons.

The difference in PQ in the screen captures here are all what has been encoded to the disc and the same decoder was used for all of them, which takes away the possibility of any of them being displayed at a disadvantage in comparison to the others. I have played Xmen 3 on my Samsung and noticed the same picture imperfections that are shown in these images.

out of interest, i found the SD-DVD version of Xmen 3 to also be a little soft.

Superman returns is quite soft on VC-1, If VC-1 is all we had i would be happy with it as at least its better than SD-DVD, but from what i have seen lately its a bit of a wash between codecs when they are at there best, so whats to argue.

All can produce the goods, and at times all are not up to the snuff.

IMO, the ONLY way to compare different methods, IS with the SAME film(obviously the same master), and there should be enough out there already to do this.

BTW rdjam if you think comparing different movies is fair game then you should take a look at the insiders thread. Where a few people have pointed out that The Prestige AVC looks a lot sharper than Batman Begins VC1. And of course Amirs response was that since they are different films the comparison is meaningless.

Of course, the same director, same director of photography, the same anamorphic 35MM, the same lighting style.... :rolleyes:

I had the "pleasure" of watching "Batman Begins" HD DVD on a Sony KDL-52XBR2 and noticed the softness, bordering on compression artifacts...almost smeared. Naturally the very low bitrate of the VC-1 encode (and liberal AVC bitrate of "Prestige") has absolutely *nothing* to do with this.

Interestingly enough, I have a Samsung October 2006 BD demo disc with the police/Batmobile (tank) chase..it's MPEG2 and it looks a helluva lot better than the HD DVD...interesting.

Of course, the same director, same director of photography, the same anamorphic 35MM, the same lighting style.... :rolleyes:
How many times do we have to loop through this same discussion?

How many times do we have to loop through this same discussion?


When you admit that just maybe starving bandwidth because HD DVD is limited to a total of 36Mbps and 30GB that stuffing a 2 hour plus film, a lossless audio track and all the extras of what was a separate DVD might not be a good idea and just might have something to do with the perceived softness. :rolleyes:

Kong proves my point: all available bandwidth is dedicated to the video. No lossless audio, no extras. Not to mention if it had been on Blu-ray there is plenty of room for even higher video bandwidth, lossless audio, and all the 2 disc DVD extras.

Of course, Kong also looked decent on a DVD5 screener. ;)

You also didn't address the fact that the MPEG2 clip of Batman Begins on the Samsung demo looks BETTER than the bandwidth starved VC-1 version :confused:

I got to thinking about various codec issues being discussed in these forums. We all know various complaints that people have with different video codecs and their deficiencies. MPEG-2 has horrid macroblocking with motion when their isn't enough bits thrown at it. We've all seen it on OTA and on some early Blu-ray discs.
.
I have seen macroblocking in BD releases and the demos and have always thought it was the transmission to the display that was bit starved not the transfer.

VC1 and AVC are more likely to have softening instead of blocking when they get bit starved. And this is preferable.

But softening really means a loss of detail. If there is much of it then it means you might as well have encoded the movie at a lower resolution since the detail isn't there anyway.

One way to measure detail is to actually test what is lost if you downrez a video to a lower resolution and then immediately scale it back to the original size. Then compare to see what is lost. The idea is that you WANT a lot of detail to be lost in this process. If it is not lost in going to a lower rez then it obviously wasn't there to begin with (in the encoded image).

I used this process in my old "Let's measure detail!" thread about 4 years ago, downrezzing HD caps, uprezzing again, and then calculating PSNR of the difference. This process could likely be used on hidef discs now too, now that we can easily rip them.

See my old thread in the archives discussing this process. My old examples are still out on my web site. (no new hidef DVD examples)
See: "Let's Measure Detail!" (http://archive.avsforum.com/avs-vb/showthread.php?s=&threadid=242229)

If nothing else it puts some numbers on things, comparing to what red laser DVD might optimally have produced if it had the bit rate.

- Tom

When you admit that just maybe starving bandwidth because HD DVD is limited to a total of 36Mbps and 30GB that stuffing a 2 hour plus film, a lossless audio track and all the extras of what was a separate DVD might not be a good idea and just might have something to do with the perceived softness. :rolleyes:
How many times do we have to loop THIS discussion?

How many times do we have to loop THIS discussion?

When HD DVD supporters admit this.

That BD is the better solution. That sooner the format was is over the better, before Warner cripples any more releases to compensate for HD DVD's deficiencies.

The MPEG2 version of MI:3 outdoes the VC1 version (see here (http://www.avsforum.com/avs-vb/showthread.php?t=813105)), which specifically shows macroblocking. There are a large number of releases--including almost every Paramount release--that are MPEG2 on one format and VC1 on another, despite your statement that they are "rare":

• Good Night, and Good Luck
• Kiss Kiss Bang Bang
• Training Day
• Full Metal Jacket
• Sleepy Hollow
• Aeon Flux
• Sahara
• Reds
• The Italian Job
• Nacho Libre
• World Trade Center
• We Were Soldiers
• Manchurian Candidate

I'm sure I'm missing a couple, too. I don't see why you can't post direct comparisons. Unless they're trying to sell you something, almost anyone will tell you to compare apples and apples. You're a consumer, so you're not trying to sell us anything.

Right?

Is that all the titles? mpeg2 and VC-1 of the same movie? Forgive me but can you post the complete list? I have a project in mind. Thanks a lot.

Here's an article (http://www.avguide.com/the-perfect-vision/76/bluray.php) outlining some of the codec comparisons made by Joe Kane. -- John

BTW rdjam if you think comparing different movies is fair game then you should take a look at the insiders thread. Where a few people have pointed out that The Prestige AVC looks a lot sharper than Batman Begins VC1. And of course Amirs response was that since they are different films the comparison is meaningless.
The point is that Batman Begins was soft in the theaters also, so VC1 isn't going to suddenly improve an already soft picture.

So, if Prestige is one of the better AVC titles, then you should be comparing it to the best VC1 titles, such as King Kong (unless you're chicken :) )...

You also didn't address the fact that the MPEG2 clip of Batman Begins on the Samsung demo looks BETTER than the bandwidth starved VC-1 version :confused:
Allow me, then...

Fact is, that you won't get very far trying to watch a 60 second trailer on a demo disc that has nothing to do with reality - any format could do a great demo.

But here, we have been comparing the REAL releases in each "VIDEO CODEC", regardless of which disc format (BD/HD) they are on. You keep seeming to forget the difference to go on and on about your favourite disc format.

VC1 is pretty consistently kicking butt over the other two codecs. Anyone with eyes can see that... Why not just have more BD releases in VC1? - after all, it's the main reason so many people still prefer HD DVD...

Eyes aren't used in science too? ;) The entire science of A/V is to bring a more pleasing presentation to the end user. So it's no surprise that if someone finds an image more pleasing, they're inclined to prefer that one, no?Did you turn on the blue stretch, green enhancement, and dynamic contrast features? :)

Allow me, then...

Fact is, that you won't get very far trying to watch a 60 second trailer on a demo disc that has nothing to do with reality - any format could do a great demo.

It's NOT a trailer. It's the actual Batmobile chase from the middle of the film.


VC1 is pretty consistently kicking butt over the other two codecs. Anyone with eyes can see that...

Then why is some of the best ever reviewed HD material out there is in AVC? Why are some of the loudest complaints (Superman Returns, The Departed, even Batman Begins) VC-1? Why does MPEG2 Paramount releases look identical to their VC-1 counterparts?

Why not just have more BD releases in VC1? - after all, it's the main reason so many people still prefer HD DVD...

No, it's cheap(er) players and some titles. And shortsightedness. :rolleyes:

It's NOT a trailer. It's the actual Batmobile chase from the middle of the film.You are diverting from the point as usual. It is still a trailer/demo scene. It is not the whole movie, so you won't get very far with the plot if you choose to watch it instead of the real movie. So sitting there and trying to compare it is a false premise. And we can only take your word for it that it looked great anyway, so for all we know it could look like poop - why not post a screenshot, then?

As I said, compare the REAL movies that are actually being sold in each format or codec.

Then why is some of the best ever reviewed HD material out there is in AVC? News to me - care to name them?

Why are some of the loudest complaints (Superman Returns, The Departed, even Batman Begins) VC-1? Says you - The Departed was VERY sharp, and had a superb picture overall.

Why does MPEG2 Paramount releases look identical to their VC-1 counterparts?Now this one is easy to prove false - stay tuned for some in depth comparison on this - I hope you like blocking artifacts... a lot :)

No, it's cheap(er) players and some titles. And shortsightedness. :rolleyes:Actually, it's due to FAR SUPERIOR players, with networking, advanced interactive HDi features (that work), decoding of DD+ and TrueHD, etc etc etc

...Oh and they HAPPEN to cost half the price of the generally rubbish, obsolete players available for BD. So we like HD DVD players because they are better, yet a smarter purchase as a bonus.

As for "shortsightedness", we'll see about that. I personally think it was VERY shortsighted of certain BD studios to have swallowed the BD "party line" and refusing to use VC1 due to POLITICS, when consumers are clearly saying they want the best. It's why I continue to buy only HD DVD, and only get the occasional BD rental...

As I said, compare the REAL movies that are actually being sold in each format or codec.This is rich, coming from someone who has yet to make a comparison between codecs on the same titles.
News to me - care to name them?The Prestige, Casino Royale, Hitchhiker's Guide...
Now this one is easy to prove false - stay tuned for some in depth comparison on this - I hope you like blocking artifacts... a lot :)It's already been proven false--in Blu-ray's favor. The blocking artifacts were in the VC-1 version.
Oh and they HAPPEN to cost half the price of the generally rubbish, obsolete players available for BD. So we like HD DVD players because they are better, yet a smarter purchase as a bonus.And this right here proves that you can't help but inject your subjective comments into the argument. How are we supposed to take your word, when you have such an obvious bias? Why not give this task over to someone who seems more or less neutral (say, skogan) and let the observations come from him, not the guy who put up the "Petition Blu-ray studios" website?

To what degree are the actual encoders used available to the public?

I would guess that it would be a whole lot simpler to get some reference material, do encodes at different bitrates, then compare the results.

Moreover, people are just comparing still-images, when video is actually about moving pictures. I would say that a down-scaled, compressed, decompressed _video_ says a lot more about a codec than a hi-rez still image.

http://www.ee.columbia.edu/~ywang/Research/VisGenie/QualityComparison.jpg

As long as the pre-processing done by studios is a creative process and the uncompressed source isnt released, it is generally impossible to compare different releases as a judge of technology.

This is exactly the same problem as those trying to prove SACD as better than CD by listening to releases available on both. As long as the title is remastered, anything can happen...

-k

You are diverting from the point as usual. It is still a trailer/demo scene. It is not the whole movie, so you won't get very far with the plot if you choose to watch it instead of the real movie. So sitting there and trying to compare it is a false premise. And we can only take your word for it that it looked great anyway, so for all we know it could look like poop - why not post a screenshot, then?The fact that it's a demo disc still doesn't explain why that scene looks better than it does in VC-1?

Says you - The Departed was VERY sharp, and had a superb picture overall.It was pretty soft compared to The Prestige or Casino Royale. Maybe it's sharp by VC-1 standards?

I'm not a codec fanboy, but the softest looking discs I own all happen to be VC-1, so I'm still trying to figure out why this is? I'm trying to figure out if it's just a Warner Brothers issue or if's something to do with filtering prior to encode, or if it's just the codec itself. I am not convinced that VC-1 is a bad thing, but I haven't seen such issues with the AVC discs I own.

Actually, it's due to FAR SUPERIOR players, with networking, advanced interactive HDi features (that work), decoding of DD+ and TrueHD, etc etc etc

...Oh and they HAPPEN to cost half the price of the generally rubbish, obsolete players available for BD. So we like HD DVD players because they are better, yet a smarter purchase as a bonus.I see you are still stuck in 2006.

As for "shortsightedness", we'll see about that. I personally think it was VERY shortsighted of certain BD studios to have swallowed the BD "party line" and refusing to use VC1 due to POLITICS, when consumers are clearly saying they want the best. Again, I don't believe VC-1 has proven to be the best. My eyes don't lie. I'm not saying it is necessarily the fault of the codec, but those Warner VC-1 titles are soft. The one Warner title I own which is really stunning is Good Night And Good Luck which is MPEG-2.

The fact that it's a demo disc still doesn't explain why that scene looks better than it does in VC-1?

It was pretty soft compared to The Prestige or Casino Royale. Maybe it's sharp by VC-1 standards?

I'm not a codec fanboy, but the softest looking discs I own all happen to be VC-1, so I'm still trying to figure out why this is? I'm trying to figure out if it's just a Warner Brothers issue or if's something to do with filtering prior to encode, or if it's just the codec itself. I am not convinced that VC-1 is a bad thing, but I haven't seen such issues with the AVC discs I own.

I see you are still stuck in 2006.

Again, I don't believe VC-1 has proven to be the best. My eyes don't lie. I'm not saying it is necessarily the fault of the codec, but those Warner VC-1 titles are soft. The one Warner title I own which is really stunning is Good Night And Good Luck which is MPEG-2.

In my opinion, the softness problem is a problem that is a Warner problem, not a VC-1 problem. A number of Warner high def releases have had a softness problem: The Departed, Batman Begins, Poseidon, Superman Returns, Lady in the Water. There have been extensive discussions in various threads here about softness in these titles, most recently concerning The Departed, which is a particularly egregious example.

This is rich, coming from someone who has yet to make a comparison between codecs on the same titles.Oh they're coming - never fear :)

The Prestige, Casino Royale, Hitchhiker's Guide...Thre whole titles? And VC1 has had scads more titles tht were given accolades. BTW - I've seen mixed opinions on Hitchhiker's PQ.

It's already been proven false--in Blu-ray's favor. The blocking artifacts were in the VC-1 version.That's funny - someone finds one single title that has some encoding errors in it and a whole thread is born proclaiming that VC1 is terrible. And then the same folks overlook the terrible posterization and video noise in almost every frame of XMen 3 :)

[And this right here proves that you can't help but inject your subjective comments into the argument. How are we supposed to take your word, when you have such an obvious bias? What's so surprising? This is a subjective subject, seeing what looks more pleasing is not ALL science. But as far as the players go, there is absolutely no doubt. I'm just putting it more brutally honestly. You get a whole lot less with BD for a whole lot more money. Sorry if pointing out this truth makes me the wicked witch.

Why not give this task over to someone who seems more or less neutral (say, skogan) and let the observations come from him, not the guy who put up the "Petition Blu-ray studios" website?Thanks for your advice, however you'd better get used to opinions coming from anyone who'd like to express them. I'm sure censorship would be great, tho...

The fact that it's a demo disc still doesn't explain why that scene looks better than it does in VC-1? Well first off, no one has proven that it DOES look better. Until someone provides a disc for some frames to be downloaded, it'll be fanbois-a-fanbois.

Second, I think it was made clear that a demo disc can afford a fairly unlimited amount of bandwidth to a very small segment, that may be completely unrealistic for a final release. Now, again, when we have the scenes in question we can prove one or the other - but otherwise it's just talk.

I remember someone saying that the POTC demo in AVC was the best demo they had EVER seen in their life. When I finally got hold of it it was HORRIBLE, absolutely riddled with noise and artifacts.

It was pretty soft compared to The Prestige or Casino Royale. Maybe it's sharp by VC-1 standards? Says you, again. We'll put the pics up soon...

I'm not a codec fanboy, but the softest looking discs I own all happen to be VC-1, so I'm still trying to figure out why this is? Very strange, then. The softest enocodes I have seen are all in mepg2. The noisiest encodes I have seen are in AVC.

Again, I don't believe VC-1 has proven to be the best. My eyes don't lie. I'm not saying it is necessarily the fault of the codec, but those Warner VC-1 titles are soft. The one Warner title I own which is really stunning is Good Night And Good Luck which is MPEG-2.Sad that you've never given your eyes a proper say ;)

KIng Kong is probably the best encode on either format using any codec.

KIng Kong is probably the best encode on either format using any codec.

This is wrong. King Kong has many frames where quantization was jacked up to 5 and 6 with many many frames where the smallest quant was 4.

What does having macroblocks quantized like that mean to the video? Well, remember when Amir did his experiment with what areas of the picture get a pass with the inloop-filter? That experiment was with a VC-1 encode using quant index of 2.

The in loop filters and overlap smoothing stuff goes up with the amount of quantization applied. King Kong is regularly using quant of 4 for most of the video and goes up to 5 and 6 when it is in trouble.

If that same experiment was done to King Kong, and someone threw up a few frames from the tough segments, how many pixels would have been filtered?

But if the smooth filtering behavior was consistent, that would at least be a statement to the capability of that filter. Problem is : it is not. I've already explained a couple of example GOPs where the I picture was grainy and detailed while the Ps were smooth, followed by Bs that are even smoother. Just look it up yourself, why else would anyone buy the hd dvd addon if not to hook up to a PC?

The problem segments in King Kong that can be studied in detail are numerous. To pretend that it is a good encode is just plain silly especially when there are so many HD DVD encodes that are better.

The problem segments in King Kong that can be studied in detail are numerous. To pretend that it is a good encode is just plain silly especially when there are so many HD DVD encodes that are better.
Hmmm... interesting analysis - but sorry to be "silly" but aside from all the technical info, King Kong is one of the best pictures I have ever seen in Hi Def. The image is so sharp, it literally looks 3D.

Even the reviewers have said the same thing - one of them even commented that he "NEVER" uses the tern 3D because he feels it is a cliche, but that this is the first title he was forced to use the word.

Glad to know I have company in the "silly" farm, tho ;) Maybe there ARE better encodes on HD DVD, also. There are so many good ones, it's sometime hard to keep track of the rankings. But I would say that this is right up there also.

Seriously - just look at this image - I mean REALLY look at it - scroll right, look at the buildings and the rooftops. The proof is in the pudding AFAIC:

http://i166.photobucket.com/albums/u98/adzez/HD%20DVD/f4d63012.png

rdjam,

The problem in posting good stills is that even uneven transfers can have amazing stills. T5E has a few stills that look incredible, and noone is gushing about what a great transfer it is because it has problem segments.

If you want to spend time looking at what can go wrong, just look at the segment before KK skids on the pond. Pay attention to the clouds in the dark sky and the snow. If you can single step, you will see differences in individual frames.

Also single step through the dark room screening of Jack Black's movie near the beginning --- find the segments with tobacco smoke : single step though and you can see detail/smooth changes frame to frame.

Also - worth pointing out that there is a lot of CGI in King Kong, whcih may have a different characteristic to real film originated material, particularly when it comes to grain (which may or may not have been added to the CGI)

I have both HD-DVD and BluRay sources (XBox 360 with Component and VGA connections and PS3 via HDMI) feeding a full 1080p Sony Bravia (40W2000 European model) I don't have enough material to accurately compare formats and platforms - but the Casino Royale BD seems to have much more obvious grain than most HD-DVDs I have, and looked really sharp throughout, with no heavily visible artefacting (apart from 3:2 judder) at first viewing. On HD-DVD I have King Kong, Bourne Supremacy and the MI series. They all look great to someone used to DVD and SD 16:9 broadcast - but Casino Royale had the edge.

No idea what codecs were in use - though watching the Bitrate meter on the PS3 during Casino Royale, it was regularly going above 30Mbs.

It FELT sharper - but I realise this is a subjective response. Could be as much the benefit of HDMI 1:1 pixel mapping (I have softness to minimum and 1:1 mapping enabled for 1080p sources) as it is the source material.

the logic being employed here is beyond stupid. ok, the king kong captures look better than the x-men 3 capture. there are literally hundreds of factors that go into that. to attribute it exclusively and exhaustively to the codec is insane. you know, the capture from "underworld" actually looks phenomenal. should i then conclude that mpeg2 is also better than avc?

"How would you test AVC deficiencies?" Don't you just compare it to a VC-1 encode and see how it differs? ;)

Worst I have seen in person is the end of Alpha Dog when Sharon Stone is in the makeup to make her fatter, look at the grain and noise on her facial features.

~Bobby

Worst I have seen in person is the end of Alpha Dog when Sharon Stone is in the makeup to make her fatter, look at the grain and noise on her facial features.

~Bobby

Nah that was directors intent to hide the wrinkles. Sharon is no spring chicken these days :)

Worst I have seen in person is the end of Alpha Dog when Sharon Stone is in the makeup to make her fatter, look at the grain and noise on her facial features.

~Bobby

If you can see grain that means the encode works well. Grain has nothing to do with video codec. :rolleyes: :rolleyes: :rolleyes:

Then what are you waiting for? ;) By the way, doesn't that statement alone indicate a recognition that The Departed is a much sharper picture overall than Crank? :)No, he clearly isn't implying anything of the sort, he is just citing examples that VC-1 can be soft and MPEG-2 can be sharp.

I like VC-1 and AVC, I like mpeg-2 as long as it is handled right. I'm a little bit puzzled why you are so attached to VC-1 Rdjam, do you have a financial interest in it? :)

I just want the best looking movies possible and that means great source materials and excellent implementaion of whichever codec is chosen for the encode. Lossless sound is nice, too.

Ok, this thread is boring. Time to resurrect the old doom9 comparison encodes threads (ie use different encoders and settings, see who can encode the original the best).

I have some quad hdtv (3840*2160, 50fps progressive, 48bit colour) scans of various scenes, so any mpeg2/vc1 encoders up for the challenge? :-) I'll use x264, since I'm not so familiar with the Microsoft encoder and probably wouldn't be fair to vc1 as a result.

Beware, you'll need bucketloads of CPU power, and even more storage - each minute of the original files is something like 150Gbytes ;)

Ok, this thread is boring. Time to resurrect the old doom9 comparison encodes threads (ie use different encoders and settings, see who can encode the original the best).

I have some quad hdtv (3840*2160, 50fps progressive, 48bit colour) scans of various scenes, so any mpeg2/vc1 encoders up for the challenge? :-) I'll use x264, since I'm not so familiar with the Microsoft encoder and probably wouldn't be fair to vc1 as a result.

Beware, you'll need bucketloads of CPU power, and even more storage - each minute of the original files is something like 150Gbytes ;)
If it's public domain content or something that we can get a license for, I'd certainly take up your offer to do a VC-1 version of it!

First thing is probably to make a 1920x1080 8-bit 4:2:0 master of it to start with - that'll be much easier to transport. And .yuv compresses up around 2:1 with .zip.

I had 6 TB at home back in '04...

This is wrong. King Kong has many frames where quantization was jacked up to 5 and 6 with many many frames where the smallest quant was 4.
What are you analyzing with? With PEP/CineVision PSE, we have differential quantization, so per-frame quant doesn't tell the whole story - a given frame can have macroblocks with quants over quite a range.

If it's public domain content or something that we can get a license for, I'd certainly take up your offer to do a VC-1 version of it!

First thing is probably to make a 1920x1080 8-bit 4:2:0 master of it to start with - that'll be much easier to transport. And .yuv compresses up around 2:1 with .zip.

I had 6 TB at home back in '04...

Let me guess Arfster's sources ...
ftp://vqeg.its.bldrdoc.gov/HDTV/SVT_MultiFormat/2160p50_CgrLevels_Master_SVTdec05_/

ftp://vqeg.its.bldrdoc.gov/HDTV/SVT_MultiFormat/1080p50_CgrLevels_SINC_FILTER_SVTdec05_/

copyright...
ftp://vqeg.its.bldrdoc.gov/HDTV/SVT_MultiFormat/SVT_MultiFormat_v10.pdf

Ben, you've already checked them.. http://www.avsforum.com/avs-vb/showthread.php?p=10081896&&#post10081896

A 1080p, 4:2:0 master would be nice. I need 1 hole month to download the 48bit images. (I am on a shared 1Mbps adsl)

@arfster. I'm glad I meet you again :)

Dethis: aha, yes, those look familiar. I'd forgotten where they came from, just some random bunch of files on a disc - downloaded them ages ago, but trying to encode that input resolution on my then PC (P4D 3.6ghz) was pointlessly slow.


BenW: thanks, good to have someone who really knows what they're doing - my encodes can be pretty decent out of sheer practice, but I don't have any specialist knowledge or frankly any real understanding of what I'm doing :-)

Quick question though, are the above suitable for this purpose?


Anyway, bit short on time, could anyone have a whizz through that copyright pdf and see if this would be OK for our purposes? In particular, could we post the finished encode for comparison? It's not going to be verifiable otherwise.

That's what I meant. Film grain issue being lack of film grain.
That's not quite what he said

BenW: thanks, good to have someone who really knows what they're doing - my encodes can be pretty decent out of sheer practice, but I don't have any specialist knowledge or frankly any real understanding of what I'm doing :-)

Quick question though, are the above suitable for this purpose?
It wouldn't be exactly linear to film on HD optical disc. The source is 50 Hz with 1/100th of a second shutter speed. We could slow down to 48 Hz and then leave out every other frame, but that'd still have 1/96th of a second shutter instead of the standard 1/48th we'd get with film content. We could slow down from 50 to 24 to get the shutter speed right, but that'd be a slo-mo effect.

There might be something interesting to do in here, but nothing so comprehensive that others wouldn't argue that the results aren't relevant :).

Still, the instructions give some hints for how to get to a .yuv source file usable by both PEP and other compression tools. Probably someone has one lying around they could share, or you could run through the instructions if you have the source already downloaed.

Anyway, bit short on time, could anyone have a whizz through that copyright pdf and see if this would be OK for our purposes? In particular, could we post the finished encode for comparison? It's not going to be verifiable otherwise.
IANAL, but I don't see anything prohibitive in the copyright terms.

It wouldn't be exactly linear to film on HD optical disc. The source is 50 Hz with 1/100th of a second shutter speed. We could slow down to 48 Hz and then leave out every other frame, but that'd still have 1/96th of a second shutter instead of the standard 1/48th we'd get with film content. We could slow down from 50 to 24 to get the shutter speed right, but that'd be a slo-mo effect.

I think slowing it to 24p sort of gets the frame rate but still probably wouldn't change the shutter speed and introduce motion blur if that is what you were looking for.



There might be something interesting to do in here, but nothing so comprehensive that others wouldn't argue that the results aren't relevant :).

Still, the instructions give some hints for how to get to a .yuv source file usable by both PEP and other compression tools. Probably someone has one lying around they could share, or you could run through the instructions if you have the source already downloaed.


IANAL, but I don't see anything prohibitive in the copyright terms.

A couple weeks ago I made a (mostly untested quickie) modified version of the Avisynth Rawsource filter that could read a sequence of these 16 bit SVT files and spit out 8 bit RGB. I could post it if anyone wanted. And of course Avisynth can then do a decent job of scaling and color conversion to whatever.

- Tom

edit: and my own vote would really to be more true to the master, not the medium. We are all playing these on a PC anyway so I'd vote to encode 1080p / 50. Most of us probably can't play that at full speed but it would still be nice to look at and compare. And at that frame rate you don't need much motion blur.

I just watched POTC2 on Blu-Ray since I heard about the stellar PQ. I got a friend who has a PS3 and he brought it over so I rented a two movies to test. It was interesting seeing the avg bitrate of POTC hover around the 12-18 range in many scenes. Some scenes were around 25-35. The lowest low was 6.5 and the highest high was 50... Which seemed strange since the bandwidth A + V on Blu-Ray according to Wikipedia is 48 with the raw data transfer rate of ~54, 50 mbps seems incorrectly displayed, I'll post pics later.

Leaving aside the possibility that the PS3 is displaying the raw disc rate( or simply getting it wrong!), there's a buffer at work. Basically stuff displayed is not read directly from disc, it goes from disc to buffer, and from there to display. Thus for veryveryvery short periods of time (0.6 seconds springs to mind), the bitrate can spike above the maximum. Of course, after such a spike the buffer is going to be kinda low, so when the bitrate next drops below the max it'll fill up again.