Official Low GOP topic

Butt

I use this setting - no problems

kae

@randolfo

"You say that we can begin to make final setting at least 24p on? With either 12 or 3 GOP is subjective, I'm still evaluating. But what about bit rates and other things. What do you say?"

I haven't tried raising the bitrate on my settings higher yet, but I think someone (you?) set it to 60 and it worked. I haven't messed with 720p or 1080i modes at all yet as I am waiting for others for input. 1080P 24 was the only thing important to me. Today I am going to try and find the upper possible limit for GOP 3. The higher the data rate the better, the cost of SD cards is not a worry for me.

kae

@proaudio4

Just change 1080i50 and 1080p24 GOP Size=3 to 1080i50 and 1080p24 GOP Size=6 and it should work fine. Sage's video looks great. I was using 6 GOP for my initial tests but like the 3 GOP better. But it's a very subtle difference. I did notice that autofocus on the 14-140 was a little snappier with the 6 GOP. And remember, it's hard to get a feel for this unless you watch it raw, so shoot your own tests with 3 and 6 and see what you like.

kae

@svart

Yes, I've transcoded many 3 GOP frames into Avid DNXHD-175 and they transcode flawlessly.

@ALL
As far as motion capture is concerned, I think of it this way - when we're shooting slow mo on film we don't shoot 24 frames and slow it down, we shoot it at 48-300 fps or higher and slow that down to get as much motion info as possible in each frame. I liken that to a shorter GOP, though I can't explain it technically. I think the shorter GOP captures more high quality motion info so on playback you 'perceive' it differently and my 'perception' of it reminds me more of film. I am still waiting for sunshine here so I can shoot another test and download the mts. I think you have to see the actual footage raw to get a feel for the difference. And you may not perceive it. I would suggest to everyone that you just flash your camera with the 1080 24p settings I posted above, shoot anything with it, but especially heavy motion, hand held preferably ("24" style) or a dolly, whatever, and watch it on a good monitor to see if you like it. Then flash back to what you had. This is the only way to settle this debate. On another note, I remember the early debate about video vs film when some posited 'hey, we can just shoot 60i video now because that's how we're watching films on television anyway.' Well, of course it didn't work because it's all in how you INITIALLY capture the image. 4:2:2 looks different than 4:2:0 for color and I truly believe short GOP looks different than long GOP for motion. Is it going to hurt the look of your movie to shoot a 12 GOP? Of course not, it'll look fine. To me, it'll just look better at a GOP of 3.

@LPowell
Thanks for finally posting a scientific explanation of this short GOP phenomena. I really respect your work. I come from a background of watching film dailies for the past 30 years so my perception of motion on film is probably different than most who simply watch heavily compressed video of films. That's why I respond to a short GOP look but I totally get it's unimportant to others.

proaudio4

Kae, Thanks for your 3 GOP information. Along with the Canogn XL2 and JVC HD100 (6 GOP), I always felt there was something more film like from motion. The JVC HD100 had it, but it also "seemed" like it had a bit of softness to the motion - probably due to the lower bit rate and overall handling of the codec parameters.

I am going to try your latest 3 GOP on the GH2 tonight, but I have heard of some using a 6 GOP on the GH2 such as Sage. Here's Sage's vimeo post of 42mbps 6 GOP:

I believe I heard the 6 GOP does not cause the first frames glitch problem?

If I wanted to use 6 GOP with your settings, what else do I need to change or set for this to work correctly?

Thank you!

cbrandin

LPowell, What do you think the overall effect would be if motion vectors were disabled and P and B frames could only use delta and intraframe macroblocks?

Chris

LPowell

As I personally have limited hands-on experience in shooting with film, I don't really have an informed opinion on what looks filmic. However, there is a perceptible difference between the look of intentional 24p motion-blur, versus the low-quality smeared artifacts produced by an AVCHD encoder when it fails to track rapid motions.

Long-GOP encoders work very well in commercial DVD and Blu-ray content production. Professional encoders provide multi-pass options that enable the encoder to accurately predict the bitrate requirements of each GOP in advance. For real-time video capture, however, a long-GOP encoder must work at an extreme disadvantage, simply because it literally cannot predict the future. That limitation forces the AVCHD encoder in the GH2 to restrict itself to conservative encoding strategies that are guaranteed to produce correctly encoded frames in the limited amount of time available to encode each GOP. With an adequately boosted bitrate, dramatically reducing the GOP-size can improve image quality by simplifying the real-time motion capture challenges that are presented to the encoder.

Elenion

This is very interesting but...then... a short gop codec really has a more filmic motion? And what's the price in terms of bitrate? I mean, in a common environment what should be the bitrate compared with a normal 12 gop?
I'm sorry for my questions but i need to clarify it on my mind, just to make better testing (and better videos eheh). Thank you!

LPowell

@svart asked: "Lets say that if you force increased I frame number with higher quality, that the P/B frames would then be forced to be of lower quality than they would normally be and thus the net sum is similar quality?"

While this encoding strategy is theoretically possible, in practice, the AVCHD encoders in the GH1/GH2 appear to have built-in mechanisms that automatically reduce I-frame size whenever P or B-frames require more bitrate. Short-GOP patches avoid this pitfall by reducing the total amount of motion in each GOP, which consequently reduces the long-duration accumulation of motion vectors that produce increased P and B-frame sizes in long-GOP encoders.

svart

Ok, I get what you are saying, thanks for explaining it.

However, would the reverse not be true as well, since you have a finite bitrate? Lets say that if you force increased I frame number with higher quality, that the P/B frames would then be forced to be of lower quality than they would normally be and thus the net sum is similar quality?

Or is the low GOP simply to keep the same net quality but only create more filmic motion?

Maybe it's just because I've been watching these test videos and haven't done any testing on this myself but I've not seen a drastic difference in the motion. After watching dozens of hours of GH1 and GH2 files while editing, I don't think the stock motion looks bad at all. Of course, I'm also transcoding to true 24p I frame-only files too.

Has anybody tried transcoding a short-GOP file? How did it turn out? I can't edit .MTS files natively and have to transcode.

LPowell

@svart Transcoding obviously can't change the motion-tracking performance of the encoder used to capture the original footage. With largely static scenes, an AVCHD encoder will generate GOP's containing large, high-quality I-frames followed by a sequence of highly compressed B and/or P-frames. When it has to deal with rapid motion, however, the encoder must divert increased amounts of bitrate to B and P-frames, and is then forced to reduce the quality of its I-frames. With a long-GOP encoder, accumulated motion vectors can often produce B and P-frames that exceed the size of I-frames, with both inefficient and low quality results.

A short-GOP encoder has an inherent advantage in motion tracking, as there simply isn't as much movement in the duration of each GOP. A 3-frame GOP (IPP) turned out to be very stable and efficient on the GH1 in SH 720p mode patched to 25/30p. On the GH2, it appears that a 3-frame or 6-frame GOP (IBB or IBBPBB) can work efficiently and produce high-quality I-frames using standard frame rates in both 720p and 1080p video modes.

Butt

thanks - i think the GOP3 this is a way
my tests see very good

randolfo

@kae

You say that we can begin to make final setting at least 24p on? With either 12 or 3 GOP is subjective, I'm still evaluating. But what about bit rates and other things. What do you say?

Aleksandr_Oleynik

@Vitaliy_Kiselev, May be 6-GOP is a compromise?

Aleksandr_Oleynik

@kae "Any ideas where I can host a 500mb file?"

Yes! - http://www.ex.ua

svart

So someone explain to me why a low GOP .MTS file would look any different than a transcoded 24p file? If you use something like cineform, you can create a file with nothing but I frames.

Vitaliy_Kiselev

I am really glad to see peaceful discussion here.
If you like low GOP it is just perfect, people are different and opinions are different.

Good advice - use GOP sizes that , if multiplied by integer value will give you original GOP.
Like GOP 3 and GOP 6 for GOP 12 modes (24p and 50i).

kae

@bwhitz

I'm sure all the codec techies will yell heresy and flash their 'BS+1s' at each other all night, but I agree with you. I first noticed the filmic motion difference when I bought my Canon XL-H1. I had used a Canon XL2 for years in 24p Native mode and its DV codec which was SD but it was INTRAFRAME. The motion looked incredibly filmic, especially with the 14X manual lens. When I saw the 24P footage on the XL-H1 I was disappointed with the motion. Then I realized why. Canon started using the HDV codec on the H1 and it was interframe. I don't why or how the GOP length affects the look and I really don't care. I'm using these 3GOP settings for all the 24p work on the GH2 until someone comes up with better ones, size of the data be damned. I ingest all clips into the Avid at DNXHD-175 anyway so the data's always going to get much bigger regardless. Would love it if someone could get rid at the glitch at the top of each shot though.

bwhitz

@kae
Yep. Definitely, looks much more like film motion.. I've been suspecting that GOP length has an effect on motion for a while now. Again I don't know all the exact details as I'm not an engineer, but it looks to me, visually, like the long-GOP quantization kills the subtle motions that give film and higher-end digital cinema cameras that "organic" and random look. Long GOP detail seems to "shift" from frame to frame instead of refreshing with the shutter. Which is why I say long-GOP has a "dual shutter speed" effect, where the majority of the motion reflects the given shutter speed (i.e. 1/50) but the rest looks more like 1/40 or 1/30.

kae

@randolfo

Just shot a PAL 1080i clip - you're right! - looks and plays perfect but of course it will have a 3 GOP if you set the 1080 24P mode to have a 3 GOP. I actually like the motion better with the 3 GOP on PAL 1080i but you don't really need it as with 50 fields motion cap is better naturally. Bet it would help if we could ever do 25p though! Thanks, I'll amend my initial post to reflect your discovery.

kae

@bwhitz
"I've been seeing an average of 40-60mbps... haven't hit 70 yet. Still getting the initial glitch, but perfect after that. Motion looks great as usual!"

Hi bwhitz,

Glad it's working for you. Doesn't the motion mimic what we see on film better at this 3 GOP or am I just imagining it??????

randolfo

Yes !

kae

@randolfo

Great news. I wonder why mine stuttered, I'll have to reshoot a pal clip and see.

randolfo

@kae
1080 50i

kae

@brudney

Okay. Will try one of those sites. And the difference isn't massively obvious, it's subtle, but subtle enough for me to want to use it rather than a higher GOP. I will say that autofocus with the 14-140 is snappier with a 6 GOP setting for 1080 24p and 6 still looks better than the 12 gop setting to me, but just on motion.

Howdy, Stranger!

Categories

Tags in Topic

Top Posters