Truthfulness of short GOP superiority claims

mpgxsvcd

@Vitaliy

I can't test it so you are right I don't know. It was just a hypothesis. So you think removing the P frames has little affect? What about b-frames? Do those have a significant affect in your opinion? Do you think the GH1 without B frames with a GOP of 3 would meet or exceed the GH2 with B frames and a GOP of 3 at the same bit rate if everything else is the same?

stoney

My unscientific, full of biased, interpretation of this topic.

It's like a vaccination. It is a noble idea... only if vaccines are made good.

How do we know that Panasonic implemented AVCHD codec that produces good B & P frames for consumer cameras like GH2?

It seems 24p looks great even at 32Mbps. GH2VK can handle higher bitrate. Why not make more I-frames? It's not breaking anything.

dkitsov

I guess is the time to reveal the sources of the footage. Most of you chose clip B as a 3GOP, clip A as a 6GOP and a clip C as a 12GOP. One of you thought that all of the clips are the same footage and only one of you were almost right.
Congratulations @bleach551 Clip A is indeed a 3GOP and clip B is indeed a 12GOP, clip C was also GOP12.
There was really no GOP6 in the lineup.
So, 4 out of 6 who bothered to vote have deemed clip B to be a 3GOP when it was a 12GOP. Only one of the 6 have correctly IDed the A as being 3GOP.
Fun.

LPowell

@mpgxsvcd
B-frames can use backward as well as forward motion vectors, which usually makes them more efficient than the forward-only P-frames. I did most of my 3-frame GOP research on the GH1, which lacks B-frames. With long-GOP encoding on the GH1, I saw many streams where rapid motion forced the encoder to pack more bits into the P-frames than into the I-frames. This causes a downward spiral in image quality, since poorly encoded I-frames provide low-quality motion vectors that require more corrective detail when reused in P-frames. That, in turn, starves the next I-frame of bitrate, which exacerbates the problem.

As long as you've got adequate bitrate to encode high-quality I-frames and P-frames, the IBBPBBPBBPBB frame-cadence of the GH2's stock 24p encoder will produce high-quality B-frames. To a large degree, the IPPIPP P-frame cadence I devised for the GH1 (Fast Action 3-Frame GOP Patch) works by mimicking the dynamic behavior of the IBBPBB B-frame cadence of the GH2.

With a 3-frame GOP on the GH2, replacing the P-frames in the stock cadence with I-frames guarantees that all B-frames will have high-quality reference frames nearby (given adequate bitrate). As Vitaliy explained above, reducing GOP-length provokes the encoder to increase its bitrate. If the increased bitrate exceeds the amount needed to replace the P-frames with I-frames, additional bits may be available to improve overall image quality. My hunch is that a high-bitrate 6-frame GOP may hit the optimal sweet spot on the GH2.

chip

Now that you say it I can clearly see the difference ;)
(No, I still can't)

stoney

Wasting bitrate don't break anything. Less compression is non-showstopper. Hard drive is getting cheaper. SDHC is getting cheaper. What's the issue here?

mpgxsvcd

@LPowell

Thanks for that response. Are you testing the GH2 now? What settings are you using right now?

stoney

@LPowell GOP6 sounds good... until stable overall bitrate breaks out 65Mbps limit. e.g. 80Mbps... or even 100Mbps.

mpgxsvcd

@stonebat

Stability, low bit rate frames at the beginning, irregular cadence, possibly bit starved frames, and playback issues. If none of those are problems for you then crank it up.

chip

>Wasting bitrate don't break anything. Less compression is non-showstopper. Hard drive is getting cheaper. SDHC is getting cheaper. What's the issue here?

Agree. I'm using kae's 65Mbit because I need to color grade most of the time. I'm using GH2 on jobs so higher quality is more important to me than better codec efficiancy, and I can definately tell it improved a lot for CC and grading. I think this thread was more about a 'filmic motion' of 3-GOP vs long GOP?! I actually still don't see that thesis destroyed by the spaghetti test

LPowell

I'm currently testing kae's 3-GOP patches to try to determine what the reliable bitrate limits are on the GH2. It's more difficult to pin down on the GH2 than it was with the GH1, which would break quite predictably. ;-)

stoney

@mpgxsvcd +1

This is sounding more and more like CPU overclocking. Some going extreme. Some playing safe. Both sides contributing to overall development.

For each PTools release, both stable and extreme settings should give more hints to improve PTools. Then we get a pair of Short GOP settings. Extreme Short GOP (ESG) and Stable Short GOP (SSG).

stoney

@LPowell I tried @kae's 65Mbps. Whenever bitrate spikes above 70Mbps, it couldn't playback unless rebooted. But not vice versa.

stoney

We say Long GOP and Low GOP.

Shouldn't it be Short GOP?

chip

>Whenever bitrate spikes above 70Mbps, it couldn't playback unless rebooted.

what card are you using stonebat? I took several clips with average bitrate at 75-76Mbit/s and they play back without error in camera (Transcend class 10)

LPowell

Yes, "short-GOP" is the term I've seen used in encoder technical documentation.

stoney

@chip Sandisk Extreme 16GB Class10

driftwood

On all the tests i did today, without any adjustments, I cant help but notice there seems to be more light in the darker areas with Kaes 65 GOP 3. The main thing I like about Kae's settings are recording with movement - yes the 3 GOP gives staccato but it just looks so filmic. With the high bitrate, the quality of the pictures seems compensated. I really need this confirming from others...

mpgxsvcd

@stonebat

That is my hope. However, ideally we will find extreme settings that have no adverse affects. Then we could have the best of all worlds.

mpgxsvcd

@driftwood

Can you post screen shots of what you are seeing with the light and darker areas.

proaudio4

@mpgxsvcd
I wished you included a high bitrate 65Mb/s 3-GOP, or 60Mb/s 6-GOP with your metronome frame grabs.
This would of been intereting to see if there was an improvement under those same conditions.

I'm currently using Kae's settings, but at 60Mb/s 6-GOP. It looks great! But I need to do testing this weekend!

APXmusic

@proaudio4 I have my camera set to Kae's settings but modified them to 50Mb/s 6GOP. Some testing that I have done looks pretty damn good!

towi

"I have my camera set to Kae's settings but modified them to 50Mb/s 6GOP. Some testing that I have done looks pretty damn good!"

Same here. For a week I've been using a setting with 52Mbits @ 24p and 32Mbits @ 1080i50 with GOP6 (and 720p50 set to GOP5) and it works without issues. In camera playback is troublefree, audio runs in sync... everything seems to run just fine so far.
While Kae's 65Mbits/GOP3 setting is working as good on my camera at 24p it doesn't work at 50i as it shows lots of artefacts.
By now I haven't discovered some really obvious advantages of 3GOP/65Mbits over 6GOP/52Mbits... but maybe there are some. This is why I'll stick with my GOP6 setting for the time being.

mpgxsvcd

@proaudio4

I couldn't test everything. I wasn't seeing much difference between 32 mb/sec and 42 mb/sec so I didn't try 65 mb/sec. I have repeated the test with 32 mb/sec and 65 mb/sec GOP=3 and GOP=6 with a continuously moving frame.

KAE's 65 mb/sec produces the highest possible bit rate of all of the settings I tested. However, all 3 samples looked the same and the 720p and 1080p Low videos were messed up for the 65 mb/sec settings. There simply weren't any compression artifacts in any of them. I can't post the samples right now because I am on a low speed connection. I will try to post them when I get back.

proaudio4

Thanks mpgxsvcd.
Looking forward to your examples.

Hopefully Vitaliy and Chris can find a way to increase the lowest bitrate; there by increasing average bitrate under the same image conditions.

Howdy, Stranger!

Categories

Tags in Topic

Top Posters