LOW LIGHT TESTS

3 SETTING FILES:

1. STOCK
2. 66M AQ2 CHRIS AUG. 29th
3. 88M AQ2 MAX VAR STRAY

All conditions are constant:
Light: Daylight Lamps
f-stop: f/5.6
ISO 320
Shutter: 1/250
stock 42
Camera reports 3 stops under exposed
Motion tests were done on a motorized tripod base.
The speed control for motion was fixed and the same for all.
_________________________________________________

Stock:
min i-frame: 420K
max i-frame: 476K
ave p-frame: 174K
ave b-frame: 47K

Images attached below:
stock static.png
stock Motion.png
stock chart.png

_________________________________________________

66M AQ2:
min i-frame: 786K
max i-frame: 876K
ave p-frame: 480K
ave b-frame: 250K

66 AQ2 static.png
66 AQ2 motion.png
66M AQ2 chart.png

_________________________________________________

88M AQ2:
min i-frame: 801K
max i-frame: 943K
ave p-frame: 533K
ave b-frame: 287K

88 AQ2 static.png
88 AQ2 motion.png
88M AQ2 chart correct image.png (Note ignore the first 88M A2 image. It's incorrect.)

__________________________________________________

SUMMARY AND FINAL THREE WAY ANALYSIS IMAGE (STATIC)

Based on the amount of detail, low light images can suffer.
Detail scenes such as landscape with trees and grass will ask for a higher bitrate
to resolve the complex fine detail. Shadow detail will be the first to go which causes detail loss
and smearing from macroblocking in the shadows.
I chose to shoot this test on my bass traps front cloth. The cloth has "some" fine detail
which forces a higher bitrate. Granted, the bitrate is not at it highest limits, but is reasonably
high enough to show differneces between the stock 22Mb/s settings compared to 66M and 88M.
Both 66M and 88M are real close in this example. The stock setting falls short here as you can see.

Check the ALL THREE COMPARE.png static image file for a quick three way comparison.

@_gl
"I personally need a balance between file size and quality. But of course I would like the best possible quality at that sweet spot. "

We all are really. The results of running the camera at extremes, and the settings used to achieve those extremes, directly inform what settings we use to get the size/quality and motion rendering quality we personally want. In order to find the correlations between the settings to get what we want you need to be able to analyse a dataset thats as wide as possible. So, seeing peoples results at extreme settings, and reading peoples theories as to the results (hinting towards what this codec is actually doing) is the only way to get that personal sweet spot. I think in the end we will get to the point where we'll know exactly what settings we will need for the AQ, GOP and bitrate we want without having to run so many tests. I do think personally that giving the codec its head so to speak, as per the max variation approach of Chris, is the correct way forward. But discoveries could be made that completely change my opinion, and they will more than likely come out of some extreme test and/or bitrate.

Whatever @stonebat, clearly not possible having an intelligent conversation with you.

Let me clarify my point as you seem to be missing it:

You're trying to get the best possible image quality, that's great. But running the codec at (say) 1/2000 shutter speed is forcing it to dedicate huge amounts of bitrate to motion, as every frame is essentially unblurred and different. My point is that if you tune the codec for that scenario _for image quality_, it will not necessarily give you the best IQ for the more common case where motion is blurred. Or it may use more bitrate for motion than is necessary for the more common 'some motion blur' case.

So sure, find settings for both scenarios (and of course reliability is crucial too), just don't use high shutter speeds as the only way to measure codec motion IQ.

And transcoding is a separate issue - the point is the GH2 codec is not intra, yes you can tune it to work more like that (lower GOP etc) but not everybody needs or wants that. I personally need a balance between file size and quality. But of course I would like the best possible quality at that sweet spot.

@stonebat, I've only done one shoot with them & haven't edited that fully yet, so it's too early to say. But from what I've seen here so far, I think the sweet spot is a general purpose bitrate that gives nearly top quality at a reasonable file size (I guess 44AQ4 or 66AQ2-3). If that could span, even better.

Of course the more extreme settings will be useful too, but I personally wouldn't use them unless I really needed to as I don't see a huge gain.

@proaudio4, @stonebat, I wasn't talking about reliability, I was talking about comparing high-shutter _motion_ with HDMI for IQ. That's (most of the time) not a real-world case.

@_gl
Well I certainly disgaree with 100% of what you just wrote. ;)

The camera should actually be able to record without failing using different settings on a camera, ISO, shutter speed. etc...
If a new unhacked GH2 failed for recording at any setting, this would be called defective. This is "real world".

Also, many do shoot at very high shutter speeds either due to:
1. They want the look.
2. They don't own an ND filter and want to maintain f-stop for DOF.
3. They shoot fast action sports

And most important why we are doing this is to find the best quality capable of running in the camera without failing.
Stray's last 88MB AQ2 MAX VARIATION (take-off from Chris's 66M MAX VARIATION version) passed my tests.
I'm in the process of posting this info now.

@driftwood Let me know when you have new GOP1 set. Appreciate your work! Would also want to test your 25P you are working on.

driftwood,
Please point me to your 3 GOP 720P settings?

I think it is a long night for @driftwood :)

anybody got any solutions for reordering frames for FCPX?

@daimon I ve been using fcp7/cinema tools to correct the frame reorder. But premiere pro 'interpret' works too. Had no probs with NeoScene too on the Mac. Playback wise, its great on the PC so long as you get a decent codec such as CoreCodec's CoreAVC(R) Professional Edition a superb H.264 Video Codec. Link: http://corecodec.com/products/coreavc

No, I havent managed (or tried again recently, tbh I can't really spend any time testing/patching the camera for a few days now) to get a stable 88M AQ2 in the same manner as Chris's original 66M AQ2. However, I think the key to doing this would be to dramatically pull down the framelimit setting, as driftwoods findings seem to suggest that an overly large framelimit setting is responsible for cadence issues. I've definitely seen doing so improve the stability. Also, my own instinct is that the high top/bottom low top/bottom also need to be cranked up a bit (like x4.8). I have absolutely no idea why now but certain results gave me the impression that doing so contributed to stability in the max variation version.

I reckon, (just by looking at the footage shot, and the bitrates its recording at), that the 66M max variation running at AQ3 is as good as the original 66M AQ2 in quality terms. I'd therefore think that (if it is as stable as it seems to be) that running this max variation 88M at AQ3 will be as good as a stable 88M AQ2 working in the original, less variable QP style.