Bit rate investigation

Started by Audionut, July 19, 2012, 04:54:03 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

JasonATL

Using 1% latest build...

Here are some captures from my NLE of some test shots from today. It wasn't windy, so most of these shots are from near static video.

I shot with three settings all at CBR 3.0x
GOP = 1 (ALL-I, right?)
GOP = 2
ML v2.3

I'm purposefully not telling which panel is which until below. So, if you want to guess, don't read lower than the link until after you have looked at the pictures.

https://picasaweb.google.com/102645356447404039003/ML_GOP_BR?authuser=0&authkey=Gv1sRgCIuEydWZ1ZXPsAE&feat=directlink




The "adjusted" are graded with sharpness (quite a bit, actually) and contrast added, plus a 1 pixel color blur added ahead of the sharpness (which serves to really cut down on the aliasing on these particular shots). You'll see in the ones with "bad" in the title that the middle one (GOP=1) had bitrate dropped (presumably when buffer was filling?) during the clip. In VLC, the bitrate here drops to about 30Mb/s. The left pane is the ML v2.3 and the right pane is GOP=2. Frankly, I don't see a huge difference. In the zoom shots, it appears that the GOP=2 might have a slight detail advantage over GOP=1. In all cases, I prefer the subtle improved look of GOP=1 or GOP=2 than v2.3, except of course for the few frames (very visible when playing it) in which the bit rate drops. The fact that I can sharpen these so much without them looking like crap really shows how much the increased bitrate helps.

Stupidly, I didn't think to try CBR 1.0x. I already know that CBR 3.0x (or 2.3x) is an improvement over Canon FW. So, I'm really trying to see what we have going here with these new improvements. Too early to conclude much, except that they are very promising.

Using a Sandisk Extreme (45 Mb/s) card, my bitrates according to Bitrate viewer were:
GOP=1 (avg ~160 Mb/s, peak around 179 varied from 75 Mb/s to 179 Mb/s throughout)
GOP=2 (avg ~149 Mb/s, peak around 152)
ML v2.3 (avg ~76 Mb/s, peak around 94)

1%

For me gop 3 seemed to perform a little better than 2 at 24p. How visible are the drops when playing?

Andy600

I've only tried GOP3 and ALL-I so far. ALL-I seems better for moving images while a GOP3 setting has better detail for static shots (I think that was to be expected anyway). I've uploaded a bunch of frame grabs from the footage I shot yesterday both ungraded and with a basic LUT applied to my LOG PS (I'm doing the real grade today). These shots were from shooting ALL-I.  http://wtrns.fr/LYPZI3ca2c1fkAj

FilmMan - That 3 second clip you have was part of a huge file. I'm not sure how to trim it to keep the bit rate etc identical and I'm on a pretty slow internet connection. I will upload some untouched files that I shot yesterday once I've finished the grade and edit for my LOG PS demo movie :)

Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

xaled

Hi 1%,

thank you for your time and efford!

there is the reference h.264 codec implementation
http://iphome.hhi.de/suehring/tml/ with a good manual http://iphome.hhi.de/suehring/tml/JM%20Reference%20Software%20Manual%20%28JVT-AE010%29.pdf, concerning the codec parameters. It might be helpful in understanding the H.264 parameters and finding them in the canon implementation.

Rush

ALL-I shouldn't be considered as best possible and best looking. Longer GOP with same bitrate should look better, because of more efficient compression mode.
Seems like 1% have found optimal GOP=3 with "highest quality + no buffer overruns + less bitrate"

GH2 have a lot of bitrate-hacks and the latest GH2 hack (Apocalypse Now) sets GOP to 3 or 6 for best quality+bitrate, so we should consider it as the best possible too...

But will 600D handle high bitrate with GOP=6 or should we stick to GOP=3?
Greetings from Russia!

deleted.account

Andy600, a quick look at your LOG PS stills, they look good, apart from no sky detail, but assume that was an exposure choice. Looking forward to hearing more about your PS and seeing some more.

What NLE and LUT format?

Andy600

Quote from: y3llow on September 12, 2012, 11:52:54 AM
Andy600, a quick look at your LOG PS stills, they look good, apart from no sky detail, but assume that was an exposure choice. Looking forward to hearing more about your PS and seeing some more.

What NLE and LUT format?

Thanks :) The shot with no sky detail was facing the sun with no ND on the lens and not really any cloud detail to see anyway. Not the easiest thing to shoot and not really a good example of the PS. There is actually more highlight detail recovered with this LOG PS than Cinestyle in my tests and without the ugly artifacts of something like Cineplus Lightform PS. I made the basic LUT using LutBuddy in After Effects CS6 and saved it as a cube file but I'll be able to provide others. It's still a little too saturated but hey it's only a LUT.

Anyway, I'll keep this topic free of any more PS talk and let you know more info in another topic when there's something to see. It's just that I was testing it at the same time as 1%'s GOP build so made sense to mention it when I show test shots.
Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com


JasonATL

Quote from: 1% on September 12, 2012, 05:33:35 AM
For me gop 3 seemed to perform a little better than 2 at 24p. How visible are the drops when playing?

Thanks. I also had GOP=3. I'll put it up on my timeline against 1 and 2 and take a look.

The drops are visible. To the casual observer, possibly not. But, of course, that's not who we're talking about here. I've posted a 5 second clip (even compressed at an average of 50 Mb/s) that you can download from Vimeo: http://vimeo.com/49305199. The drop happens right around 0:01 and you can see what I understand to be macroblocking. Would this change if I changed DblockA and DblockB?

In looking at one of my GOP=3 clips, I notice the same thing. Less noticeable in terms of the resolution loss, but clearly something "digital" happens at certain points.

Is this is a result of dropping the bitrate (and QScale) to keep the buffer from overflowing or is this something on the Canon side? I thought I understood the buffer to not come into play for GOP=1, so I wonder if it is something in the Canon algorithm. I shot a total of 30 clips and had a sheet on which I took notes of settings for each shot. But, I didn't take detailed notes of what I saw on the debug screen. I do recall seeing only a few clips in which changes were clear, but I didn't mark them. More testing this weekend...

Let me know if none of this is helpful. I'll continue testing for myself, but I don't want to muck up this thread with my observations if they aren't on point.

Andy600

Jason ATL - If you change the deblocking filter to 0/0 the macro blocking should reduce. I'm experimenting with -1/-1 ATM which seems to give a slight increase in the perception of sharpness/detail and add a little fine grain. I think it's more of an aesthetic thing.
Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

Andy600

Jason - If you look at the .mov in Bitrate viewer you can see exactly where on the timeline the bitrate drops and then pixel peep the video
Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

FilmMan

Quote from: Andy600 on September 12, 2012, 06:23:06 AM
I've only tried GOP3 and ALL-I so far. ALL-I seems better for moving images while a GOP3 setting has better detail for static shots (I think that was to be expected anyway). I've uploaded a bunch of frame grabs from the footage I shot yesterday both ungraded and with a basic LUT applied to my LOG PS (I'm doing the real grade today). These shots were from shooting ALL-I.  http://wtrns.fr/LYPZI3ca2c1fkAj

FilmMan - That 3 second clip you have was part of a huge file. I'm not sure how to trim it to keep the bit rate etc identical and I'm on a pretty slow internet connection. I will upload some untouched files that I shot yesterday once I've finished the grade and edit for my LOG PS demo movie :)

Andy600,
No problem.  Only post if time permits.  I know it takes alot of time in what your doing.  I appreaciate all your efforts and posts.  The plane shots are awesome by the way.  Cheers.

1%

If its blocking, deblocking filter would take care of it. I can't casually tell on my videos because I haven't found a subject where it's visible.

You can set gop to whatever you want up to 100 something. Your  buffer is the limiting factor here. I think after 3 or 6 its diminishing returns, the camera only handles a certain amount of predicted frames at a time.

The dropping function still needs some work... maybe it needs to drop less. Not sure... could make the bottom limit a setting. I was mainly experimenting with it to stop overflows. Without it camera couldn't handle filming static... with it,  holds up most of the time. If its a a big noticeable drop during grading/watching then there is a problem.

Wonder how well the 4:2:2 mjpeg holds up with the buffer. Wish a1ex would release it even if not done or perfect. Sounds like it will be a while since he is working on 5DMKIII pretty heavily.

JasonATL

Quote from: Andy600 on September 12, 2012, 02:33:22 PM
Jason - If you look at the .mov in Bitrate viewer you can see exactly where on the timeline the bitrate drops and then pixel peep the video

Thanks, Andy600. I edited the post above to have Bitviewer stats and will use those from now on.

Andy600

Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

AriLG

Is the GOP thing available to 550D users, too ? or is it exclusively a 600D thing (for the time being) ?
T3i (main), T2i
------------------
It's not about accuracy,  it's about Aesthetics

nanomad

It Will probably work on any camera but you will probably be stuck with a low video duration
EOS 1100D | EOS 650 (No, I didn't forget the D) | Ye Olde Canon EF Lenses ('87): 50 f/1.8 - 28 f/2.8 - 70-210 f/4 | EF-S 18-55 f/3.5-5.6 | Metz 36 AF-5

jgharding

Quote from: nanomad on September 14, 2012, 04:09:42 PM
It Will probably work on any camera but you will probably be stuck with a low video duration

So will this be permanent, or do other bodies like 550D just need some specific code?
Zeiss primes, 600D, a lot of shadow. http://www.jgharding.com

nanomad

Best case: you need to find the same functions in your camera fw
Worst case: you'll need to start from scratch.
550d is probably in the middle
EOS 1100D | EOS 650 (No, I didn't forget the D) | Ye Olde Canon EF Lenses ('87): 50 f/1.8 - 28 f/2.8 - 70-210 f/4 | EF-S 18-55 f/3.5-5.6 | Metz 36 AF-5

jgharding

Ah well, I wanted to get a 600D for the flip out screen anyway ;)
Zeiss primes, 600D, a lot of shadow. http://www.jgharding.com

Andy600

Quote from: jgharding on September 14, 2012, 05:06:26 PM
Ah well, I wanted to get a 600D for the flip out screen anyway ;)

Don't get it for the flipout screen. Get it for the sensor crop ability ;)
Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

AriLG

Quote from: Andy600 on September 14, 2012, 05:11:06 PM
Don't get it for the flipout screen. Get it for the sensor crop ability ;)
So, what you are saying is that if I shoot with my - say - Tokina 11-16, I'm effectively shooting REAL HD at 33-48 (with the X3 Crop) ?
T3i (main), T2i
------------------
It's not about accuracy,  it's about Aesthetics

nanomad

Yes, and with no moire effect too  :o

Actually, I'd like a test on one little thing as it was a subject of discussion between me and Bart :P
The 600D is an APSC camera, so it has a "default" zoom of 1.6x
Crop mode adds 3.3x zoom.

So, let's say I've to a 200 EF lens, does this become 200x1.6x3.3 = 1000 mm in crop mode?  :o

It should be easy to test by comparing a zoom lens with a short fixed one:
1. EF 28mm 3x crop vs 150mm of the 18-200 EF-S non-crop
2.  60mm of the 18-200 EF-S 3x crop vs 200 mm of the same lens
EOS 1100D | EOS 650 (No, I didn't forget the D) | Ye Olde Canon EF Lenses ('87): 50 f/1.8 - 28 f/2.8 - 70-210 f/4 | EF-S 18-55 f/3.5-5.6 | Metz 36 AF-5

jgharding

Hmm that's maths o'clock. I'd say 1056mm (relative to 135 full frame), as it's 200mm * (1.6*3.3)... give it a test!

It's a huge crop, good for super tele! I'd have thought the noise grains will be much larger. Given the deeper DOF aesthetic it could be useful for certain shots with that Tokina.

If you're used to thinking in Super 35 or APS-C anyway (as I am), its 660mm.

Everyone talks in terms of the 1.6 crop for APS-C, but that's only relevant if you think in terms of photographic full frame. Standard movie film has always been a different world (in general), it's only recently that Super 35 has been talked of as cropped! I think of 135 frame (5D etc.) as oversized...  :o

Zeiss primes, 600D, a lot of shadow. http://www.jgharding.com

nanomad

My google-fu just demonstrated that's it actually like that, you're getting a focal length of X*3.3*1.6 if you're using EF lenses
On the 550D you can even get  a *7.2*1.6 magnification if you shoot in VGA mode. Now, that's some kick-ass zoom :P

600D / 200mm / 3X crop


550D comparison footage
EOS 1100D | EOS 650 (No, I didn't forget the D) | Ye Olde Canon EF Lenses ('87): 50 f/1.8 - 28 f/2.8 - 70-210 f/4 | EF-S 18-55 f/3.5-5.6 | Metz 36 AF-5