Quality of H.264 and H.263

[Spoon]

Recently I did some experiments using a pair of Tanberg 880 units. Running them over IP at a range of data rates (64k to 384k) we swopped between H.264 and H.263. To say I was rather surprised with the results is an understatement.

H.264 was significantly lower quality !!!! Very blocky with poor motion handling and lots of artifacts.

Can anyone explain to me what the reason for this is ? Was I doing something wrong ? Does the 880 unit not really have the horsepower to cope with H.264? or perhaps this H.264 thing is just a marketing gimmick ?

Not had the chance to try a VSX-7000 in the same way. Has any one else ?

Perhaps I'm just not bright enough to notice how much better it is

Thoughts ??

[Shaft_323]

This post was supposed to be for you but I guess I pressed new topic. Anyway......

Interesting, were you comparing apples to apples (AKA 384 using H.264 to 384 using H.263) or were you comparing H.263 vs. the equivelent H.264 line rates (AKA H.263 @ 384 vs H.264 @ 192K)
Did you take any actual baseline measurements of bandwidth utilization during these tests?

Thanks for sharing!

[jont]

Guys I have also done similar tests with the 880 and found the results to be equally poor. The tests were conducted between two 880s at 128/192/384/512/768

Each test began using H.264 with the person at the far end moving around quite a lot, then I changed the endpoint to H.263 without dropping the call. In my opinion it was markedly better in H.263 than in H.264.

Now given that Tandberg have been touting "much improved video quality" I would be interested to hear shauns (our resident Tandberg rep) response to this and what can be done to improve the call quality.

Jon

[tom9933]

The pair of VSX 7000 units we have here seem to work very nicely using H.264. I have been very pleased with the quality and was pleasantly surprised when doing calls at extremely low bitrates (i.e. 56, 64 and 128k). What I have noticed is that the breakup looks different with H.264.
Tom Mills
Wright State University

[Keisuke Hashimoto]

As far as I saw a demo of H.263 vs H.264 on VSX7000 for the first time, my impression was that H.264 was better than H.263. I am not sure if it was "markedly" or not, but I felt that it was better but it might be influenced by my previous notion that H.264 was better than H.263 before I actually saw the difference.

How you feel or get impressed is subjective depending on each person so it may vary slightly.

[Shaft_323]

Keisuke

Was the comparrison apples to apples (AKA: 384K to 384K) or was it more real world? (like 384K to 192K)?

At 384 H.264 should look a good deal better than H.263 but at 192K h.264 should look at least as good as H.263 (or H.261).

H.263 was designed to provide better video than H.261 at bandwidths below 192K, anything above that should be compared to H.261 & H.263 as I have seen many systems at those and higher line rates where the H.261 looks just as good or better than the H.263

[Keisuke Hashimoto]

Shaft_323,

As for the bandwith used, 384kbps was used but the demo was showing a DVD music that showed a rock band playing some music with dark background or dim light.

That might be why I could not notice much and it was a quick demo anyway. Maybe I have to have another chance in which I try that in office environment or better test environment.

Other than that, I thought that the equipment was very good and the market would accept this new model in a favorable way.


<FONT COLOR="#000000" SIZE="1" FACE="Verdana, Arial">This message has been edited by Keisuke Hashimoto on 30 October 2003</font>

[Mero]

Shaft_323:H.263 was designed to provide better video than H.261 at bandwidths below 192K, anything above that should be compared to H.261 & H.263 as I have seen many systems at those and higher line rates where the H.261 looks just as good or better than the H.263

Shaft I agree and like to add a little math:
The difference between H.263 and H.261 depend on the content of the scene. The H.263 algorithm introduces an additional "object shape" assumption about which area on the frame does contain moving parts and which does not. It is assumed, that
one big egg-shaped object is moving before a still background. This egg mainly varies at the borders and in the lower third.If the real content of the video stream matches this assumption, the compression of H.263 is better than H.261 In other words: A single person sitting before the cam takes advantage, if the illumination is good enough for avoiding cam noise flickering on the background. Although this matches a single person view of a proper PTZed group system too, the advantage of H.263 comes up mainly in desktop videoconferencing. As this historically is a lower bandwidth application, the legend of H.263 was born to be optimized for low-bandwidth.
It should be better read as: Non-H.263-matching scenarios are typically those having high bandwidths installed.

The truth inside the legend is the simple issue, that visible compression artifacts are much more easy to generate at lower bandwidth, while higher bandwidth requires more sophistical material like e.g. a doctor´s view on a small detail of severe medical interest. But this is valid for all codecs including H.264 and MPEG.

I see H.264 to be a step towards a general video algorithm able to scale to all realtime and recorded applications from lowend webvideo up to HDTV broadcast. In the future, this will increase the demand for a single type of codec chips and therefore lower the price for VTC hardware. Actually H.264 does increase the costs by demanding higher processor power and therefore new hardware investment. Í suggest to welcome H.264 if one likes to invest anyway, but not to throw away hardware because of H.264 only.

Mero

[Shaft_323]

The promise of H.264 is that it is an encoding scheme that allows better video encoding at lower line rates. H.264 is aimed squarely at SOHO markets where Cable modems and DSL are the norm (AKA: Not a lot of bandwidth to play with) however the net effect is that larger customers can utilize this encoding scheme to reduce the bandwidth taken up by video traffic on their netoworks or reduce ISDN charges OR allow them to get better performance out of the same infrastructure.

Comparrisons should be made with this in mind and if you are using high bandwidth already you would want to compare this to your existing calls to see if it provides any benefit.

[Shogun2]

We have recently implemented Tandberg 2500's on 128K - 384K ISDN for a government client.
My impression of the 128K H.264 was that the video appeared to be less pixilated at the lower bit rate than H.263, and stabilised the image quicker. The client was very impressed with the image quality, having previously had VC's between the sites on older equipment.

Rod

[Mero]

Rod, thanks for the report

Shogun2 My impression of the 128K H.264 was that the video appeared to be less pixilated at the lower bit rate than H.263, and stabilised the image quicker.

I please you and other contributors to add a short description of what kind of scene and movement you played with to get the results.

To make subjective impressions more dependable, it is needed to add two infos to a H.264-H.263 comparison at same bandwidth:

• Was the video observed a single person´s head or another scene (see above)?
• Is the framerate comparable on both tries or is lowering of pixelation payed by lowering the rate?

See my post about the hidden ratio parameter all H.26x codecs have inside. Playing with this ratio can manipulate the impression of two different codecs. Tandberg was "high rate" playing in the past. Maybe they now shifted to more "high accuracy" ...

In our test lab we generally use recorded video sequences to reproduce exactly. Developers of compression algorithms use standardized sequences like "Travelling Salesman" to make results comparable at different locations.

For a more "independent" comparison, I suggest to camcord at least two very different scenes at customer´s typical usage and replay this during demonstrations. This makes it possible to look on framerates and on block artifacts sequentially.

Mero

<FONT COLOR="#000000" SIZE="1" FACE="Verdana, Arial">This message has been edited by Mero on 31 October 2003</font>

[Shogun2]

Mero,

This was not a definitive test by any means.
We had me on one end with an upper torso shot, and a group at the other end - I was doing the initial client training.
I was quite animated, as I am normally during a training session, and the group had different peeople moving constantly.
Most of our conferencing is at 128K due to the high cost of bandwidth, so I am used to the "normal" quality of the H.263. The H.264 was superior, as I said, as the pixilation was definitely less pronounced.
We also ran duo video at 128K, with Powerpoint and a talking head successfully. This is something I would not attempt normally with H.263 at 128K.

Rod

[Shaft_323]

The fact that H.264 looks better than H.263 at 128Kbps is not interesting for me. It makes perfect sense that h.264 should look much better than H.263 when comparred at the same line rate becuase the encoding algorythm's are much more efficent.

The test I would like to see is a comparrison of the following: (I wish I had the equipment to do this)

Using a DVD player to produce motion video!

H.264 @ 128Kbps vs. H.261 @ 384
H.264 @ 128Kbps vs. H.263 @ 384
H.264 @ 768Kbps vs. H.261 @ T1 (1536kbps)
H.264 @ 768Kbps vs. H.263 @ T1 (1536kbps)

The H.264 should look equal to or better than the H.261 or H.263 singals.

Also, monitoring the network and record actual bandwidth usage for each call.

If anyone has done this or has a chance to do so please post your response.

[miked]

Who cares!!! The bottom line is does the customer like it and does it do what they want. What customer is going to go through all of that testing and looking a bandwith utilization before they buy. What if H.264 at 128k only looks as good as H.263 at 192k instead of 256k, will that be a failed sale? I doubt it. I stated a few months back that all the hype of H.264 was just like the hype of H.263 and all of the rest of the standards. Every time some new standard came out everyone said that videoconferencing would be saved. It seems that not much has happenned. VC companies are still going out of business and margins are still not good. Surely it will turn around, but not because of H.264 but because the economy will change. Polycom introducing the VSX-7000 is a good first step but I have already seen it advertised for just over $4000. Who can make any money with that?

[Shogun2]

I have to agree with you, Mike.

Most of my clients are technically illiterate, and don't care about H.264/3/0 etc., as long as it provides a reasonable VC experience.

All we have done is raise the bar a little bit further in terms of quality, and in this case it means that they will purchase more units.

One thing that does concern me is the reported incompatibility between the Tandberg and Polycom implementations of H.264.
Can anyone shed some light on when this is to be "repaired"? Afterall, this is supposed to be an ITU standard.

Rod