On2 brags about their encoder vs x264.. [Archive]

LoRd_MuldeR

27th August 2009, 13:42

Yeah, but I want a trial-version for testing (not buying), especially since I got my lovely x264 :)

I assume they won't give it away for free (not even a trial version), because somebody could make a "real" test against state-of-the-art H.264 encoders.

At the same time people that already payed big money for VP8, will favor it anyway, because they must legitimate their investment ;)

BTW: "On2 proprietary compression solutions bring unique commercial advantages."

I guess that statement found on the On2 web-site applies to the developer/vendor, not to the user/customer :p

Forteen88

27th August 2009, 13:45

"On2 proprietary compression solutions bring unique commercial advantages."

I guess that statement of On2 applies to the developers, not to the user/customer :pAha, Ok, seems like they don't have a version for consumers, just a SDK.

Gokumon

27th August 2009, 18:06

This just in. Company tries to make itself look better than its competition. More on this breaking story at 9 o'clock.

Dark Shikari

27th August 2009, 18:11

IMO One2 don't use the best available H264 encoder. Anyway VP8 seem a really powerfull codec. VP8 seem on par with x264 at max quality for OPSNR ...And Mainconcept can beat x264 too in PSNR; that's not news.

IMO, if it cannot even beat x264 significantly in terms of PSNR, VP8 is a massive disappointment; x264's psy optimizations surely put it far ahead of VP8, so if VP8 cannot even beat x264 without psy optimizations, we cannot expect it to be remotely useful in reality.

It's been 6 years since H.264 was released--they had all this time to write their own specification from scratch--and they still fail this badly? Even I was expecting better than this. What a joke.

Edit: It's worse than I thought; they left AQ on for the x264 test, which suggests that VP8 will probably lose by >1db PSNR in reality. Nice job being completely irrelevant, On2.

smok3

27th August 2009, 18:28

brag in what way?
that vp8 is blurry?
that they have to use vp6 to show comparision between vp8 and x?

this looks like a complete anticommercial for them;

mediainfo vp8x264_vga.flv
General
Complete name : vp8x264_vga.flv
Format : Flash Video
File size : 15.8 MiB
Duration : 1mn 11s
Overall bit rate : 1 844 Kbps
Encoded date : Fri Sep 12 13:18:33 2008
Tagging application : On2 Flix SDK for Dshow_2.4.2.2
canseektoend : No

Video
Format : VP6
Duration : 1mn 11s
Bit rate : 3 272 Kbps
Width : 640 pixels
Height : 480 pixels
Display aspect ratio : 4:3
Frame rate : 29.970 fps
Bits/(Pixel*Frame) : 0.355
Stream size : 28.0 MiB

CruNcher

27th August 2009, 18:45

that whole thing their is their old almost 1 year ago Introduction page i see no new info the Video is also as old so im not sure why to discuss it again, in the H.264 section :P

see http://forum.doom9.org/showthread.php?t=141107

Sagittaire

28th August 2009, 00:44

And Mainconcept can beat x264 too in PSNR; that's not news.

Mainconcept at max OPSNR quality never beat x264 at max OPSNR quality in all my test for many source. As far I know x264 is the best available H264 codec in the area for OPSNR result and I make test with many H264 encoder.

IMO, if it cannot even beat x264 significantly in terms of PSNR, VP8 is a massive disappointment; x264's psy optimizations surely put it far ahead of VP8, so if VP8 cannot even beat x264 without psy optimizations, we cannot expect it to be remotely useful in reality.

VP8 seem to have advanced psy optimisation too ...

It's been 6 years since H.264 was released--they had all this time to write their own specification from scratch--and they still fail this badly? Even I was expecting better than this. What a joke.

I don't think. One2 is really serious company. In the past all the annonced result was correct for VPx. Anyway it's always easy to choose good source for VP8 codec.

Edit: It's worse than I thought; they left AQ on for the x264 test, which suggests that VP8 will probably lose by >1db PSNR in reality. Nice job being completely irrelevant, On2

It's true one2 don't use OPSNR tuned setting for x264. Anyway with max OPSNR quality x264 is just in par with the annonced result for VP8 from One2.

Dark Shikari

28th August 2009, 00:51

VP8 seem to have advanced psy optimisation too ...Then why did VP7 look so bad if On2 is supposedly good at psy optimizations? It fell victim to the "blurring is good" RD fallacy worse than any other modern encoder I have ever tested, completely destroying detail across a wide variety of sources.

G_M_C

28th August 2009, 08:43

What i seaid before; I think that whole page is NOT aimed at really comparing differences. It's only disigned as a marketing tool. It says more about their potential clients than it does about the encoding software itself.

From this i can deduce that their potetial customers are at least sort of gullible, since they fall for this kind of misleading marketing. Also you can deduce that they dont know how to use Google, else they would try to find out what this 'x264' they mention actually is. Not that that would help, i think they would not know how to compare encoding software for them selves for that matter.

And when you take all this into account; VP8 is good for people like that, let them; Imagine the flooding of n00b-questions if they they actually did start to use x264 :D

Anyway, in short: The comarison on that page is not intended to be fair, it's just intended to mislead potential customers in to buying that crap.

benwaggoner

30th August 2009, 06:19

Then why did VP7 look so bad if On2 is supposedly good at psy optimizations? It fell victim to the "blurring is good" RD fallacy worse than any other modern encoder I have ever tested, completely destroying detail across a wide variety of sources.
That's really been a hallmark of On2 codecs going back to VP3, and was certainly a big part of VP6.

And that's probably not entirely wrong from a consumer expectation perspective; people notice added wrong stuff (blocking and ringing) than they notice right stuff missing (loss of detail). Blocking is particularly HVS attractive due to all those nice horizontal and vertical edges, which are about the most interesting thing to the human eye. Perhaps should have taken a lesson from Sony's ClearVID patterns and rotated our blocks 45 degrees, or from our misspent wargaming youths and used hexagons :).

Even semi-pro compressionists can often get into the trap of doing QA by looking for artifacts, and assuming that if there's no blocking and ringing, the encode is going to be "good enough." But that's not a good assumption with a modern codec, or an advanced decoder with postprocessing. Comparing a VP6/VP7 AVI with different postprocessing settings in the DirectShow decoder can be very revealing.

Dark Shikari

30th August 2009, 06:24

That's really been a hallmark of On2 codecs going back to VP3, and was certainly a big part of VP6.

And that's probably not entirely wrong from a consumer expectation perspective; people notice added wrong stuff (blocking and ringing) than they notice right stuff missing (loss of detail). Blocking is particularly HVS attractive due to all those nice horizontal and vertical edges, which are about the most interesting thing to the human eye. Perhaps should have taken a lesson from Sony's ClearVID patterns and rotated our blocks 45 degrees, or from our misspent wargaming youths and used hexagons :).This would explain why Microsoft's psy optimizations are such a failure: this entire paragraph is completely wrong.

Blurring is horrific HVS-wise; blocking and ringing are vastly preferable. Compressionists easily spot them, but what ordinary users will spot is the complete destruction of detail throughout the video. Artifacts are far more preferable than destruction of detail, especially as the former tends to mask the latter. Blocking is only noticeable in flat areas of the frame; in high-complexity areas it actually improves perceived quality in many cases, especially at low bitrates and in high motion, where the blocks appear for too short a time to be seen, but for long enough to create perceived sharpness.

The human eye loves sharpness and the appearance of detail. In particular, just like the human ear, it wants to see energy conserved. The heavy blurring of encoders like VP7 is the exact opposite of this, which is why it has such a terrible ratio of PSNR to actual visual quality. The same problem exists with wavelets; they destroy perceptual detail and sharpness.

(Should I really be helping our competitors this much? ;) )

benwaggoner

30th August 2009, 06:37

This would explain why Microsoft's psy optimizations are such a failure: this entire paragraph is completely wrong.
Well, bear in mind I didn't play much a hand myself in the psychovisual optimizations in any Microsoft codecs. I've pointed out issues in a number of real-world samples in the last few years, but I can't claim much credit or blame there.

Blurring is horrific HVS-wise; blocking and ringing are vastly preferable. Compressionists easily spot them, but what ordinary users will spot is the complete destruction of detail throughout the video. Artifacts are far more preferable than destruction of detail, especially as the former tends to mask the latter. Blocking is only noticeable in flat areas of the frame; in high-complexity areas it actually improves perceived quality in many cases, especially at low bitrates.
I'm in fact speaking about cases where blocking is visible, due to high quantization in flat/smooth regions. When high texture detail masks block boundaries, there's no problem.

Up to a point, a little blocking and ringing can produce some perceptual sharpness that can mask some loss of detail. But once it's easily detectible by viewers, I think it's more actively distracting than a slight loss of detail; the real question is how much artifacting versus how much blurring is the right balance at a give level of compression.

I'm thinking of MPEG-2 as the canonical example here; the Canopus MPEG-2 encoder does a better psychovisual job by preventing blocking, even though detail may not be preserved as well.

The human eye loves sharpness and the appearance of detail. In particular, just like the human ear, it wants to see energy conserved. The heavy blurring of encoders like VP7 is the exact opposite of this, which is why it has such a terrible ratio of PSNR to actual visual quality. The same problem exists with wavelets; they destroy perceptual detail and sharpness.
I think we're actually in agreement on the salient points here.

VP7 looks a lot better viewed without reference than in an A/B comparison with the source, or with another encoder. And this goes back to VP3 at least; a lot of what Xiph has done has been to fix tables and such to allow high detail when sufficient bitrates are available.

(Should I really be helping our competitors this much? ;) )
x264 is open source; do you even have competitors :)?

Dark Shikari

30th August 2009, 06:41

I'm in fact speaking about cases where blocking is visible, due to high quantization in flat/smooth regions. When high texture detail masks block boundaries, there's no problem.That's not related to blocking or ringing, that's a failure to use proper adaptive quantization.I'm thinking of MPEG-2 as the canonical example here; the Canopus MPEG-2 encoder does a better psychovisual job by preventing blocking, even though detail may not be preserved as well.MPEG-2 doesn't have a loop filter. Of course, VC-1 doesn't have a useful loop filter either.
x264 is open source; do you even have competitors :)?Yes, our competitors are the entire commercial world of encoders who wish we didn't exist.

Recent blind tests by a major French broadcaster put x264 ahead of the best commercial broadcast encoders, despite the testing being interlaced and x264 not supporting MBAFF.

There are very few applications where x264 isn't a competitor.

benwaggoner

30th August 2009, 06:58

That's not related to blocking or ringing, that's a failure to use proper adaptive quantization.
Which is a relatively new feature of codecs. I'm taking an historical view here.

MPEG-2 doesn't have a loop filter. Of course, VC-1 doesn't have a useful loop filter either.
Well, it's still better than nothing :). And really, the In-Loop Deblocking and Overlap Transform filters should really be considered as two halves of VC-1's in-loop postprocessing; the loop filter itself isn't the whole story and wasn't designed to be used in isolation. But yes, I'd say that the H.264 in-loop filter is probably its single most impressive and impactful feature.

Yes, our competitors are the entire commercial world of encoders who wish we didn't exist.

Recent blind tests by a major French broadcaster put x264 ahead of the best commercial broadcast encoders, despite the testing being interlaced and x264 not supporting MBAFF.
Congrats. That's quite impressive given the lack of MBAFF.

There are very few applications where x264 isn't a competitor.
Well, x264 certainly competes with those encoders, but do they really compete with x264? You don't really lose anything if someone uses another encoder. x264 doesn't require commercial competitors to lose in order to "win" on its own terms.

Even though Microsoft has its own H.264 encoder, I don't view x264 as a competive threat to anything we're doing, but as another complementary part of the H.264 ecosystem that we also participate in.

Dark Shikari

30th August 2009, 07:04

Which is a relatively new feature of codecs. I'm taking an historical view here.MPEG-2 encoders have had adaptive quantization for years now. Good ones have, at least.Even though Microsoft has its own H.264 encoder, I don't view x264 as a competive threat to anything we're doing, but as another complementary part of the H.264 ecosystem that we also participate in.That's because you aren't trying to sell your encoder; it's not a significant part of the business model. You merely need to have an encoder. In fact, if you dropped it today and used x264 wherever the license made it possible, you wouldn't lose much of anything. You wouldn't even be admitting defeat, since your goal was never to produce one of the world's best encoders to begin with.

By comparison, Mainconcept can't do such a thing, because their business is selling an encoder.Well, x264 certainly competes with those encoders, but do they really compete with x264? You don't really lose anything if someone uses another encoder. x264 doesn't require commercial competitors to lose in order to "win" on its own terms.Perhaps this legend might be relevant here.

One day in the pavilion at Karakorum Genghis Kahn asked an officer of the Mongol guard what, in all the world, could bring the greatest happiness.

"The open steppe, a clear day, and a swift horse under you," responded the officer after a little thought, "and a falcon on your wrist to start up hares."

"Nay," responded the Kahn, "to crush your enemies, to see them fall at your feet -- to take their horses and goods and hear the lamentation of their women. That is best."

Victory is nothing without the defeated.

benwaggoner

30th August 2009, 07:42

That's because you aren't trying to sell your encoder; it's not a significant part of the business model. You merely need to have an encoder. In fact, if you dropped it today and used x264 wherever the license made it possible, you wouldn't lose much of anything. You wouldn't even be admitting defeat, since your goal was never to produce one of the world's best encoders to begin with.
Well, our goal certainly is to produce one of the best video encoding and publishing products. But the codec implementation itself is a relatively small part of that. Most compression tools fail at source decode, preprocessing, or workflow before they ever get to the codec itself.

I hope we'll get MPEG-4 passthrough working in a future version so x264 encoded content can be used with Expression Encoder to do trimming, metadata insertion, and player generation without recompression.

By comparison, Mainconcept can't do such a thing, because their business is selling an encoder.
True. It's interesting how small a part that pure codecs play in the overall compression tools business. In terms of revenue and headcount, codec development is well less than 10%.

Perhaps this legend might be relevant here.

One day in the pavilion at Karakorum Genghis Kahn asked an officer of the Mongol guard what, in all the world, could bring the greatest happiness.

"The open steppe, a clear day, and a swift horse under you," responded the officer after a little thought, "and a falcon on your wrist to start up hares."

"Nay," responded the Kahn, "to crush your enemies, to see them fall at your feet -- to take their horses and goods and hear the lamentation of their women. That is best."

Victory is nothing without the defeated.
Exactly.

You need to be better than Main Concept for you to win.

Main Concept needs people to buy licenses for their stuff anyway for them to win.

And everyone's happy :)!

creamyhorror

30th August 2009, 08:05

Exactly.

You need to be better than Main Concept for you to win.

Main Concept needs people to buy licenses for their stuff anyway for them to win.

And everyone's happy :)!
Actually, I think he needs to have many more commercial users than MainConcept in order to have beaten them :D But to do that x264 has to win the marketing war.

Underground78

30th August 2009, 08:48

Recent blind tests by a major French broadcaster put x264 ahead of the best commercial broadcast encoders, despite the testing being interlaced and x264 not supporting MBAFF.

I guess the name of this major French broadcaster has to be kept secret ?

Sagittaire

30th August 2009, 11:45

Blurring is horrific HVS-wise; blocking and ringing are vastly preferable.

Perhaps for your eyes but it's completely false for common users. Personnaly I don'k like blur but I don't like blocking and particulary mosquito noise too. For example DivX active by defaut PP4 for MPEG4 ASP because common user prefer decoding like that. In VP8 vs H264 demo I prefer by far VP8 bluring vs H264 blocking+ringing ...

drmpeg

30th August 2009, 12:13

Recent blind tests by a major French broadcaster put x264 ahead of the best commercial broadcast encoders, despite the testing being interlaced and x264 not supporting MBAFF.

There are very few applications where x264 isn't a competitor.
x264 competes in the broadcast encoder market? Is there a real-time encoder that takes in HD-SDI and outputs DVB-ASI that's based on x264?

Ron

Dark Shikari

30th August 2009, 15:54

x264 competes in the broadcast encoder market? Is there a real-time encoder that takes in HD-SDI and outputs DVB-ASI that's based on x264?

RonNo, but a major broadcast company sells headends which use nothing but x264 ;). So I guess the answer would actually be "yes", you just can't buy it as a separate box.Perhaps for your eyes but it's completely false for common users. Personnaly I don'k like blur but I don't like blocking and particulary mosquito noise too. For example DivX active by defaut PP4 for MPEG4 ASP because common user prefer decoding like that. In VP8 vs H264 demo I prefer by far VP8 bluring vs H264 blocking+ringing ...So you debunk my claim by saying that "common users" prefer blurring... and then back that up with... your own preferences? You're not a common user; stop trying to pretend you are one.

Feel free to keep thinking that Blur().Blur().Blur() is a great psy optimization--the rest of the world will keep moving on without you. It is people like you who are the cancer killing video quality, believing that it is a valid "optimization" to lowpass every source before encoding it because there's nothing wrong with blurring.

The fact of the matter is that the eye likes sharpness, which is why so many Blu-rays and DVDs are oversharpened, because it often simply looks better. As such, an appearance of sharpness and detail is extremely important.

Of course the whole argument is stupid because you will happily endorse any encoder out there merely because it isn't x264. If a competitor to x264 blocked more than x264 did, you would claim that blocking was good. If a competitor to x264 ringed more than x264 did, you would claim ringing was good. The opinions of shills have zero value.

CruNcher

30th August 2009, 16:55

The most annoying artifact though is happening in the restricted area (Broadcast,Blu-Ray) and that is by far "I frame pulsing" it kills any other artifact you are used too and exactly that is where VP8 i think tried to demo back then: "look same bitrate same I frame interval but we have no visual pulsing, and our Demo doesn't falls apart" (we put the bits more balanced so you see only some detail lose in the way of a blur out all frames the same) :)
At least that is what i saw in their Demo Video the artifacts you see there are happening in x264 when it reaches the bitrate resolution boarder to stay stable (Chroma falling apart, the old ASP trials,AQ is firing back edges fall apart in motion,all the motion problems with B-frames, and especially the pulsing @ the i frame change).
Though x264 since these 1 year old results was enhanced and improved especially for such situations it's much better and hits this boarder not so soon anymore (compression efficiency improved again) even in 1 pass (live encoding).

benwaggoner

30th August 2009, 16:56

Feel free to keep thinking that Blur().Blur().Blur() is a great psy optimization--the rest of the world will keep moving on without you. It is people like you who are the cancer killing video quality, believing that it is a valid "optimization" to lowpass every source before encoding it because there's nothing wrong with blurring.
I don't think he or anyone is making such an extreme statement. Everyone agrees that accurate detail is the goal; the question is what sorts of degredation are preferable when that goal can't be met.

Really, deblocking in post-processing is fundamentally a blur, but as it is blurring out erroneous detail it's much less controversial. No encoder should bake in a low-pass, but there are times where reducing detail to prevent artifacts is a good context-specific psychovisual optimization.

The fact of the matter is that the eye likes sharpness, which is why so many Blu-rays and DVDs are oversharpened, because it often simply looks better. As such, an appearance of sharpness and detail is extremely important.
I find ringing from oversharpening pretty much as annoying as ringing from overcompression :).

But getting back to On2, the VP6/7 decoder will actually synthesize noise as part of postprocessing to exactly achieve the effect you're talking about. I'm somewhere between horrified and admiring by that :), but it inarguably produces more visually pleasing results when it's on but not off.

One non-technical downside to the H.264 loop filter is that by preventing artifacts it allows less efficient encoders to get away with poor detail retention (looking at you, Apple) because the output still doesn't look "compressed" with palpable artifacts. At least with VC-1, when it's mangled the source, there's no hiding it :).

VP6 and particuarly VP3 had a pretty low maximum detail retention capability which became more obvious when postprocessing was off. I didn't spend enough time with VP7 to see if that was true there as well.

Does anyone know if VP8 continues any of that advanced postprocessing technology, or if they've moved everything in-loop?

Dark Shikari

30th August 2009, 16:59

But getting back to On2, the VP6/7 decoder will actually synthesize noise as part of postprocessing to exactly achieve the effect you're talking about. I'm somewhere between horrified and admiring by that :), but it inarguably produces more visually pleasing results when it's on but not off.I would say the primary reason for that is as a dither, which is IMO a very good idea; it reduces the need to waste bits in flat areas to avoid blocking and blurring. I would completely agree, for example, with putting gradfun2db as a required postprocessor in some future video standard.

benwaggoner

30th August 2009, 17:02

The most anoying artifact though is happening in the restricted area and that is by far I frame pulsing it kills any other artifact you know and exactly that is where VP8 i think tried to demo look same bitrate same I frame interval but we have no pulsing (therefore we put the bits balanced so you see only some detail lose) :) at least that is what i saw in their Demo Video
What do you mean by "restricted area?"

Keyframe strobing is one of those things that causes far more psychovisual pain than PSNR or SSIM would imply.

benwaggoner

30th August 2009, 17:07

I would say the primary reason for that is as a dither, which is IMO a very good idea; it reduces the need to waste bits in flat areas to avoid blocking and blurring. I would completely agree, for example, with putting gradfun2db as a required postprocessor in some future video standard.
Well, hopefully we'll get universal > 8-bit support down the road as well for the same reason; Windows 7 now supports GPUs and imaging running in all kinds of > 8-bit modes, including linear.

Encoding in 10/12-bit and having the decoder handle correct conversion to 8-bit for legacy displays would be the ideal way forward, I'd think. Heck, you could even do some display-specific tweaks, using a heaver dither for low luma on LCD for example.

3+ years out at least, though.

FWIW, zambelli did a cool demo using the Silverlight pixel shaders to add noise postprocessing with a QP-adaptive strength. Worked pretty well as a lab test, although probably not worth the extra CPU load for many users.

CruNcher

30th August 2009, 17:44

What do you mean by "restricted area?"

Keyframe strobing is one of those things that causes far more psychovisual pain than PSNR or SSIM would imply.

was still editing i meant especially areas like Broadcast and Blu-Ray with for them uncommonly low bitrates (efficiency loss), and yes it's the most painfullest things even i would prefer a whole blur out (like VP8 showed it) of everything instead of every x seconds seeing this rapid not expected detail change :(

Sagittaire

30th August 2009, 20:59

Feel free to keep thinking that Blur().Blur().Blur() is a great psy optimization--the rest of the world will keep moving on without you. It is people like you who are the cancer killing video quality, believing that it is a valid "optimization" to lowpass every source before encoding it because there's nothing wrong with blurring.

Agree or not agree common user prefer blur vs blocking or ringing. I don't understand why you use default inloop at 0,0 if for you bluring is the really most important artefact. Really popular codec like old RV10 are here for prove that.

The fact of the matter is that the eye likes sharpness, which is why so many Blu-rays and DVDs are oversharpened, because it often simply looks better. As such, an appearance of sharpness and detail is extremely important.

Well you don't know anything in bluray encoding simply because you are not compressionist. In the past the most impresive quality for common user are HDDVD (VC1 title ... sorry) and they are not really oversharpened but more too blury (for my eyes anyway).

Of course the whole argument is stupid because you will happily endorse any encoder out there merely because it isn't x264. If a competitor to x264 blocked more than x264 did, you would claim that blocking was good. If a competitor to x264 ringed more than x264 did, you would claim ringing was good. The opinions of shills have zero value.

I've had enough of your insults (for me and other). It's not because you are a x264 developer that you are always right my friends. If a moderator could remind that it would be nice. You should stop to spend most of your time in front of your computer and go out a little. Find you a girl too because you seem really to need that ...

Atak_Snajpera

30th August 2009, 21:58

I've had enough of your insults (for me and other). It's not because you are a x264 developer that you are always right my friends. If a moderator could remind that it would be nice. You should stop to spend most of your time in front of your computer and go out a little. Find you a girl too because you seem really to need that ...
What he does in spare time it is not your business. This forum is not a place for personal advices.

Sagittaire

30th August 2009, 22:09

What he does in spare time it is not your business. This forum is not a place for personal advices.

Speak about that with Dark Shikari and not with me ...

Theliel

30th August 2009, 23:36

Sagittaire, both. DS and you.

IMO, im considered me a normal user, and i prefer a lot of more a blocking that a bluriring. For me, Blurring its a detail killer

IgorC

31st August 2009, 00:15

I see two very extreme positions:
1. Blocking
2. Bluring

But what happens in the middle? Is there good balance between blur and blocking. It's somehow adjustable with -psy -deblock and maybe other options too.

Why it should be exactly 0 or 1? There is also 0.5678...... infinite combinations.

Qubit is a superposition of values: 0 and 1.

Of course, everybody heard about Schrodinger's cat. http://www.youtube.com/watch?v=7SjFJImg2Z8&fmt=18

There is no such things as absolute 0 neither 1 in nature.

plonk420

31st August 2009, 05:46

No, but a major broadcast company sells headends which use nothing but x264 ;).

heh, the company i work for uses the company that merged with your company as VOD provider :O

Revgen

31st August 2009, 06:26

I don't know if I'm a common user, but I will say this. I kept using Xvid for years instead of X264 because I simply couldn't stand the blurring. Until DS and others significantly improved the AQ, Psy, Inloop and other factors of X264 to keep it from becoming a blur beast, I wouldn't even use it.

Do I hate blocking? Sure, but not as much as I hate blurring. With blocking, my brain can at least piece together what's missing. With blurring, I sometimes wonder if I need glasses before I realize it's the actual video doing it.

nakTT

31st August 2009, 06:38

Sagittaire, both. DS and you.

IMO, im considered me a normal user, and i prefer a lot of more a blocking that a bluriring. For me, Blurring its a detail killer
For me, I prefer blurring over blocking (somewhere in the middle would be better, though). But the thing is, I get non of those by using x264 because even at 500kbps, I can get myself a very good visual quality video.

Blue_MiSfit

31st August 2009, 06:55

I prefer blocking over softening in most cases, except in extremes when it becomes impossible to overlook the blocks.

I find things like keyframe pumping and banding much more objectionable, to be honest. But, these are more easily avoided IMO, and should be handled in pre / post processing. I'm really looking forward to higher bit depth video, and more advanced post-processing. I'm getting tired of fiddling with ffdshow's noise generator! :)

And guys... keep the personal insults out of the forums. That kind of stuff doesn't belong here. Let's try to be thankful for each other's perspectives :D

~MiSfit

Dark Shikari

31st August 2009, 08:33

Find you a girl too because you seem really to need that ...Oh how so many of our competitors wish it were so... ;)I don't know if I'm a common user, but I will say this. I kept using Xvid for years instead of X264 because I simply couldn't stand the blurring. Until DS and others significantly improved the AQ, Psy, Inloop and other factors of X264 to keep it from becoming a blur beast, I wouldn't even use it.This is where my logic originally came from: I listened to people whining for days upon weeks upon months about how Xvid looked "sharper" and "more detailed" than x264, which seemed ridiculous--until I did comparisons that showed that in many cases it was true. And then I fixed x264, of course.

This is why I distrust "professional" opinions from compressionists; I have dealt for years now with the complaints of ordinary users--people who have no idea how codecs work and are just complaining about how the video looks to their eyes. And I have not once seen a complaint that x264 is "too blocky" compared to a competitor, but I have hundreds of times (pre-psy) seen complaints about detail and sharpness. In general, I'm going to trust the ordinary users, not the professional with trained (and biased) eyes.

Plus, blurring everything out is easy. Retaining detail is hard. I prefer a challenge.

Sagittaire

31st August 2009, 10:09

Then why did VP7 look so bad if On2 is supposedly good at psy optimizations? It fell victim to the "blurring is good" RD fallacy worse than any other modern encoder I have ever tested, completely destroying detail across a wide variety of sources.

http://www.on2.com/image.php?blob_id=539

Here I prefer by far the VP8 bluring over x264 blocking+ringing. I don't know if this test is realistic (certainely not good test for x264) but VP8 seem promising.

Always the same problem with you ... exaggeration !!! You speak about VP7 and other codec too like mainconcept but you are unable to make encoding with good setting for other codec. I have always better result than yours when I make myself the encoding with mainconcept for example. In the same time you always destroy the other tests because they don't use top settings for x264.

Dark Shikari

31st August 2009, 10:15

http://www.on2.com/image.php?blob_id=539Er, are you trying to prove my point or what?

Fine detail, like the woman's sunglasses and collar, are completely gone in the VP8 version. And the human eye is better able to perform temporal interpolation through artifacts than it is through blur (the latter is actually impossible), so in motion the win by x264 will be even larger.

And this is despite the fact that On2 was able to cheat in that test all they wanted (pick bad settings for x264, pick best possible test clip for VP8, cherry-pick the best possible frame, etc).

Arguing with you is hopeless; no matter what the facts say, you will instantly take a contrary opinion solely because you are paid to hate x264 and promote inferior competitors. There is no point in taking this further.

Sagittaire

31st August 2009, 10:36

microchip8

31st August 2009, 11:44

Sagittaire, I think it's you who's trolling here and this is not the first time you do so. Looking at the above pictures, I prefer the H.264 one instead of the smeared, almost out of focus blurred VP8 one. When it comes to blocking + detail retention and blurring + no blocking, I'll virtually always pick the former one unless it is really extreme in blocking, which x264 never does that with sound settings, no matter your claims. The eye is very sensitive to detail and prefers it so when you show someone a blurred pic or video and one that has all the details there but blocks not too strongly, it'll virtually always prefer the latter one. Ringing? I don't mind a bit of ringing but since I've never experienced in normal x264 encodes any ringing so far (at least if it was there, I couldn't notice it), the point your trying to make here is non-existent and most of the things you say border on trolling and bogus claims. That's all I have to say and if this conversation continues like it is now, it's a waste of time

Sagittaire

31st August 2009, 11:53

Sagittaire, I think it's you who's trolling here and this is not the first time you do so.

Seriousely ... you certainenly don't read some thread from Dark Shikari here ... speak about that with modo if you want ... lol

Looking at the above pictures, I prefer the H.264 one instead of the smeared, almost out of focus blurred VP8 one. When it comes to blocking + detail retention and blurring + no blocking, I'll virtually always pick the former one unless it is really extreme in blocking.

Seriousely ... ???
Poll in progress ... I want see if my eyes are good or not!

which x264 never does that with sound settings, no matter your claims. The eye is very sensitive to detail and prefers it so when you show someone a blurred pic or video and one that has all the details there but blocks not too strongly, it'll virtually always prefer the latter one. Ringing? I don't mind a bit of ringing but since I've never experienced in normal x264 encodes any ringing so far (at least if it was there, I couldn't notice it), the point your trying to make here is non-existent and most of the things you say border on trolling and bogus claims.

Well x264 always make massive blocking and ringing at high level quantisation like all other codec in the area I think ... no ... ???

ajp_anton

31st August 2009, 12:08

I also prefer the h264 image. Note that I do not think it looks good in any way. I just think it looks "less bad" than the VP8 one.

microchip8

31st August 2009, 12:09

Seriousely ... you certainenly don't read some thread from Dark Shikari here ... speak about that with modo if you want ... lol

Yes, seriously. I've read many threads by Dark and in not one of them I've seen him claim things which he can't back up

Seriousely ... ???
Poll in progress ... I want see if my eyes are good or not!

Yes seriously. The blurred one is so bad to my eyes that I wouldn't stand watching it for a minute. The guy's head has almost merged with the head of the woman behind him, which if it was a bit more blurred, it'll give the impression as if he surgically did something to his head or is carrying his baby twin on the back of his head. And since there's no detail that separates them apart, I wouldn't tell which one it is. His twin or the surgery he had :p

Well x264 always make massive blocking and ringing at high level quantisation like all other codec in the area I think ... no ... ???

All block and ring at high quantization levels but do you really always encode normal DVDs or Blurays at such high levels? No, unless you're an insane person who for some reason loves such things. CRF 20 has yet to produce here any noticeable and disturbing blocking or ringing, And that's what I use to encode. Not some absurdly high CRF value with high quants

Sagittaire

31st August 2009, 12:29

Yes, seriously. I've read many threads by Dark and in not one of them I've seen him claim things which he can't back up

Well you don't read the good thread ...

Yes seriously. The blurred one is so bad to my eyes that I wouldn't stand watching it for a minute. The guy's head has almost merged with the head of the woman behind him, which if it was a bit more blurred, it'll give the impression as if he surgically did something to his head or is carrying his baby twin on the back of his head. And since there's no detail that separates them apart, I wouldn't tell which one it is. His twin or the surgery he had :p

You have the complete demo video here:
http://www.on2.com/index.php?599

You prefer always the block+ring scenario vs blur scenario ... ???
No seriousely ... come on?

All block and ring at high quantization levels but do you really always encode normal DVDs or Blurays at such high levels? No, unless you're an insane person who for some reason loves such things. CRF 20 has yet to produce here any noticeable and disturbing blocking or ringing, And that's what I use to encode. Not some absurdly high CRF value with high quant

There are not only DVD and Bluray ripp scenario in the encoding area. This demo is clearly a low bitrate streaming scenario (for Youtube, Dailymotion for example).

Chengbin

31st August 2009, 12:37

This thread will get closed down if we don't stop attacking.

Sagittaire, you are entitled to your own opinion, but you have no right to force your view to someone.

Dark Shikari may "talk" as if he is a troll, but as froggy1 said, I've never seen him claim things that he can't back up.

This blur vs block argument is really based on personal perference. One thing people didn't mention is DISTANCE. When I'm reading this thread, I prefer the VP8 image, but when I sit back a meter or 2, I prefer the H.264 (I can't see blocking, but only detail)

Try to be nice to each other :)

Atak_Snajpera

31st August 2009, 12:41

Here is perfect example where so called blocking is alot better vs bluring!
http://forum.doom9.org/showthread.php?p=1314161#post1314161
Would you still prefer blured grass?

microchip8

31st August 2009, 12:45

Well you don't read the good thread ...

Trust me, I read the "good threads" and I'm sure people will agree that you're the one pushing your own view and are incapable of accepting others. The h264 one just looks better to me... period. If you call me biased or something, then we all are biased including you.

You have the complete demo video here:
http://www.on2.com/index.php?599

You prefer always the block+ring scenario vs blur scenario ... ???
No seriousely ... come on?

Come on what? Because the h264 looks better *to me* you can't accept that? I can say the same thing... Come on you preferring the blurred one...

I just don't like such an amount of blur. It's just too much, IMO

There are not only DVD and Bluray ripp scenario in the encoding area. This demo is clearly a low bitrate streaming scenario (for Youtube, Dailymotion for example).

Yeah I know that, but even here x264 can stand its man and is definitely better at low bitrates/high quants than anything I've seen so far from On2. Again, this is how I prefer things (more ringing/blocking but with detail preservation than no ringing/blocking but with massive blurring)

Chengbin

31st August 2009, 12:52

Here is perfect example where so called blocking is alot better vs bluring!
http://forum.doom9.org/showthread.php?p=1314161#post1314161
Would you still prefer blured grass?

The problem with that comparison is there is no blocks! It is a comparison of blur vs detail, which, of course we all like the more detailed encode.

Atak_Snajpera

31st August 2009, 12:58

The problem with that comparison is there is no blocks! It is a comparison of blur vs detail, which, of course we all like the more detailed encode.
Blocks are there but well hidden in grass :) You must watch whole sample.

CruNcher

31st August 2009, 13:09

Er, are you trying to prove my point or what?

Fine detail, like the woman's sunglasses and collar, are completely gone in the VP8 version. And the human eye is better able to perform temporal interpolation through artifacts than it is through blur (the latter is actually impossible), so in motion the win by x264 will be even larger.

And this is despite the fact that On2 was able to cheat in that test all they wanted (pick bad settings for x264, pick best possible test clip for VP8, cherry-pick the best possible frame, etc).

Arguing with you is hopeless; no matter what the facts say, you will instantly take a contrary opinion solely because you are paid to hate x264 and promote inferior competitors. There is no point in taking this further.

Sure the detail preservation is much better but in motion x264 fails here, because of the extreme "i-frame pulsing" that is non existent in the unified blurred VP8 example, because their is no extreme detail change @ all.
And i guess that was On2's Demo Point here (restricted low bitrate situations).Though as i said previously x264 was enhanced since that time and the boarder @ which this happens is higher now (Dark also fixed some Chroma bug and others).

So this comparison is really outdated for both x264 and On2 and can't be taken visually seriously anymore.

And yes im also one of those who prefers sharpness over blurring since the dawn of time, but i also prefer in special situations stability over detail preservation and that i was reading out of this Demo from On2 so in this very special case i would agree that On2 did a better job, but as Dark Shikari said it was a very special case that should show the weakness of x264 and some old bugs played into the hands of On2 here too.
Of course the other issue is AQ seems to backfire in such situations here as it compensates the inloop deblocking strength and becomes especially visually problematic in motion (ask the Joost Engineers they know what i mean) and that way "i-frame pulsing" "keyframe strobing" "keyframe pumping" call it however you want is going to show up heavily, personally i know this problem since a while and i can definitely say this only happens if you try to go insanely low bitrate for the given resolution as some of the psy enhancing stuff starts to backfire and also the default settings brake apart.

But most of the normal x264 users wont ever reach that especially with the default settings it's impossible (unrestricted).

nakTT

31st August 2009, 13:12

Blocks are there but well hidden in grass :) You must watch whole sample.
Agree, you must watch the clip before passing judgement.

IgorC

31st August 2009, 19:28

It's not directly blocking/blur comparison but still relevant to it. Dark Shikari has already performed the test for different values of psy at low bitrates. http://forum.doom9.org/showthread.php?t=146507

Final results (WARNING! It's low bitrate. At higher bitrates higher values of PSY are more preferble)
1. 40% of people prefered PSY 0.
2. 34% - PSY 0.2
3. 15% - PSY 0.4
4. 11% - PSY 1.0

As everybody can see there is no absolute best. Personally I think that PSY 0.1 will be more optimal (higher percentage of people's preferement) for those conditios of test.

Gentelmen, let's be more logical. That VP8 and H.264 screen has nothing to do with _particular_ deblocking or bluring filters at all. And there is no even original source image. Nothing to talk about.

Dark Shikari

31st August 2009, 20:00

It's not directly blocking/blur comparison but still relevant to it. Dark Shikari has already performed the test for different values of psy at low bitrates. http://forum.doom9.org/showthread.php?t=146507

Final results (WARNING! It's low bitrate. At higher bitrates higher values of PSY are more preferble)
1. 40% of people prefered PSY 0.
2. 34% - PSY 0.2
3. 15% - PSY 0.4
4. 11% - PSY 1.0That test was with QNS, which utterly changes the results. They're not at all comparable to what you would get with normal x264. I completely agree that QNS + high psy screwed things up hardcore ;)

Biggiesized

2nd September 2009, 02:16

Can I push here? Both have tradeoffs when it comes to artifacts. What I'd like to see is an "in-between" encode. This static shot isn't a compelling case for either camp, I'm sorry.

Revgen

2nd September 2009, 02:27

Whether On2 or X264 is better is irrelevant when it comes to the sample clip.

I can just insert blur().blur().blur() in my avisynth script and feed it to x264 if I want a result like On2 gives. Which I don't.

Like DS said, blurring is a lazy way to compress. It's easier to do than maintaining detail.

Ajax_Undone

2nd September 2009, 06:34

Agreed!!!

+1 for Schrodinger's Cat ;-)

bvc2068

5th September 2009, 01:02

Just to add another users opinion on "blurring vs. blocking": I cannot see the relevance of comparisons where either effect is as clearly visible as in the two On2 sample pictures posted above.

When I compare encoded material with the original (in motion, not as still frames, of course), and there's either a clearly visible blurring or a clearly visible blocking/ringing, then the only answer for me is to increase the bitrate.

What I consider a well-working encoder is one where I don't have to increase the bitrate to achieve visual encoding transparency in just a few scenes while the rest of the material looked fine already with a much lower bitrate. x264 does a very good job on this, the only scenes where I still see some room for improvements are very dark ones where sometimes color gradients around lights become visible at (average, 2-pass) bitrates higher than those where I could not distinguish the rest of the material from the original.

I wonder whether "youtube" has altered the perception of the younger generation such that they "expect" to see encoding artefacts, that would explain the ridiculously low bitrates used with some of todays digital media. Even what e.g. youtube calls "HD" is so much below the quality that I am willing to watch that I really suspect people must have forgotten what material without encoding artefacts looks like.

Bottom line: In my opinion the criterion for encoder comparisons should be "bitrate at which visual transparency is achieved for >50% of the people". Unlike other values, this one is even easy to measure: Just don't tell the test watchers whether they are seeing encoded or original material, let them guess for a number of scenes which ones where compressed, and increase the "visually transparent" counter for the tester if his guesses did no significantly exceed random choice probability.

Revgen

5th September 2009, 01:32

I wonder whether "youtube" has altered the perception of the younger generation such that they "expect" to see encoding artefacts

The "younger generation" watch Blu-Rays and DVD's too. They know the difference between youtube and the higher bitrate formats.

They also know that youtube is free and DVD's at the local electronics store are not.

bvc2068

6th September 2009, 17:39

They also know that youtube is free and DVD's at the local electronics store are not.Well, following that logic they would also happily feed from garbage cans and dress in rags, because those are free sources for food and clothing. But especially with non-essential goods like entertainment there usually is another choice: Instead of lowering the price by lowering the quality, you can also lower the quantity (amount of video footage that you offer or consume) and keep the quality constant.

And by the way, for music compression formats it has always been a widely accepted goal to keep the compression losses below what most people are able to perceive.

I wonder why this should be different with video compression, in a time where storage space, bandwidths and computing power are quickly getting cheaper.

Revgen

7th September 2009, 02:14

^No. Eating out of garbage cans and watching youtube have nothing in common. You're trying too hard with that one.

Free is free. So far youtube and youtube clones are the best quality available for free. For awhile DIVX had a site called Stage 6 that was way superior, but they shut it down. Until another company offers streaming video at superior quality like Stage 6 did for awhile, youtube is here to stay.

Dark Shikari

7th September 2009, 02:17

Free is free. So far youtube and youtube clones are the best quality available for free. For awhile DIVX had a site called Stage 6 that was way superior, but they shut it down. Until another company offers streaming video at superior quality like Stage 6 did for awhile, youtube is here to stay.There are thousands of Youtube clones, many of which have superior quality.

Comatose

7th September 2009, 13:52

From this i can deduce that their potetial customers are at least sort of gullible, since they fall for this kind of misleading marketing.
Like most authoring studios?

Anyway, x264 wins that comparison simply because their motto is "big videos, small files", but blurred that much it might as well be low res...

And DS... you're always so aggressive, even if the other person is apologetic (not as in this case)... reading your posts is kinda eek.

MisterH

7th September 2009, 16:57

I wonder whether "youtube" has altered the perception of the younger generation such that they "expect" to see encoding artefacts, that would explain the ridiculously low bitrates used with some of todays digital media. Even what e.g. youtube calls "HD" is so much below the quality that I am willing to watch that I really suspect people must have forgotten what material without encoding artefacts looks like.

Although the US is basically the center of the internet and generates much of it's content, many international consumers of such content don't have a wondrous fiber connection straight to the home, some countries even have download caps.

As you've said, you value compression.
I doubt just throwing more bitrate at everything and calling it a day is going to motivate many people to improve compression technology. The more demand there is for providing high quality at lower bitrates, the better. (Video conferencing started all this, right?)

Besides that, unless google plastered ads everywhere, the bandwidth cost of youtube providing proper HD would be unsustainable. There are plenty of clones out there, but I doubt they have millions of people watching and uploading videos everyday.

Sagittaire

7th September 2009, 16:59

And DS... you're always so aggressive, even if the other person is apologetic (not as in this case)... reading your posts is kinda eek.

It's true ... :devil:

Atak_Snajpera

7th September 2009, 17:21

Youtube's HD quality is not that bad
http://www.youtube.com/watch?v=PUxz11PsCnM&

Dark Shikari

7th September 2009, 17:29

Youtube's HD quality is not that bad
http://www.youtube.com/watch?v=PUxz11PsCnM&That's because they use x264 :p

Chengbin

7th September 2009, 17:38

Youtube's HD quality is not that bad
http://www.youtube.com/watch?v=PUxz11PsCnM&

Wow, that's very good quality @ 2mbps.

I never knew YouTube used x264!

JEEB

7th September 2009, 17:39

Just on the topic of Youtube 'clones' that convey better quality and that have at least some number of users:

Nicovideo (http://www.nicovideo.jp/)
Zoome (http://www.zoome.jp)

Both services take user-encoded files as well as let you upload videos and have them encoded on the server-side. Nicovideo goes up to 600/1000kbps and has certainly millions of users. Maximum resolution is around 800x600 or so, bigger resolutions were banned because of Flash's well-known "speediness" with such.

Nicovideo restricts viewing to people having an account and to certain sites that have been given rights to use the external player (even with the account restrictions nicovideo has been quite popular - mostly in Japan, but also abroad - mainly because of the amount of original content etc. spawning from the site).

Zoome, on the other hand, is smaller than nicovideo, but lets you have up to 1500kbps videos that are under 720p and up to 3000kbps videos for 720p or further (file limit is 200MB or so). Their recent partnership with Yahoo! Japan tells me they have gotten the interest of some companies, and, although they surely have less users than nicovideo, they still are not the smallest fish in the pond. Mostly known as the place where the higher quality versions of videos get put, both abroad as well as in the video-watching Japanese communities.

Zoome has no restrictions on viewing as per accounts go, but they have now started enforcing a rule of using a bittorrent framework to deliver the high-def content.

Of course, worldwide usage of these sites is smaller than Youtube's, but Nicovideo was actually fighting with Youtube for the title of the most used flash-using video site in Japan back in 2008 (interwebs should find anyone interested the needed figures).

There you have it, based on what kind of things you want to see you have the option to watch something or upload something with much better quality than Youtube allows you to. It's just that those services are not always known in your neighborhood.

And on the topic of what started this thread, I guess On2's decision is just to think that a more "stable" picture is better than having details "stand out" - although that just means you're having less details and more blur(). No matter how hard to believe, I also actually prefer the x264 picture in that certain set of pictures provided by them.

This was my bit of tl;dr.