AVCHD maximum image quality settings and testing

Ralph_B

66M AQ2 Motion

Ralph_B

88M AQ2 Static

Ralph_B

88M AQ2 Motion

stip

Crazy! Look how far you guys pushed this whole thing. We're talking about minimal differences between AVCHD and HDMI recording. VK, Chris and all guys testing and helping, you are now in my book of cool people ;)

peterdv_dot_com

@Ralph_B

Thank you for doing all those tests & posting your results. Very useful info!

cosimo_bullo

@Ralph_B - Amazing! Thank you.

stoney

@Ralph_B What a kick-ass test. So cool to use the HDMI version as reference.

cosimo_bullo

@Ralph_B - What were your camera settings for these? What ISO?

Ralph_B

Shutter speed was 1/2000 ISO 640 Lens: 14-140 @ 45mm f6.3

I may do some night tests at high ISO to see if any differences are more obvious.

cosimo_bullo

Ralph, is there any chance you could repeat the 66 AQ2 test with ISO 160. Sure you're busy, but if the opportunity arrises... Thanks again for all you're doing.

Ralph_B

@cosimo_bullo

Not sure what the point would be. The original 66 AQ2 test (one page back) was done at 1/1000, ISO 320.
What do you think we would learn that we don't already know?

cosimo_bullo

I'll take a look at the 320iso results. Probably will tell me everything I want to know. Thank you.

cosimo_bullo

Sorry to throw a monkey in the wrench here but here's a thought after our limited testing this evening. I think we may be going about this wrong (to some extent) in that we're looking for the 'best' overall settings, but not keeping in mind just how much ISO changes how a given setting performs. For example, cbrandins 66 AQ2 looks really great in the shadows at 160 ISO, but bump the ISO to 640 and the shadow detail starts to block up. At 640 ISO, eiji's very high 132 gop3 AQ2 looks better than the 66 setting. But at 160 Chris's settings look as good (if not possibly a little better) than eiji's. I guess what I'm trying to say is that maybe we should make sure we're on the same page ISO-wise as we test different settings. For example, we rarely use high ISO settings, so the goal would be to find the best settings for 160-320 ISO. Others may do a lot of night shooting and would want to tailor to 1600-3200.

Ralph, you asked about the point of testing AVCHD vs HDMI at 160 iso. The point is that I believe the GH2 codec is so much happer dealing with 160 (even vs 640) that the shadow results with AVCHD will be much stronger vs HDMI. I hope. Right now there's still a lot of break up in there, leaving me at least looking at the HDMI image with a wistful sigh.

chauncy

"Right now there's still a lot of break up in there, leaving me at least looking at the HDMI image with a wistful sigh."

There is? In the grabs here? Where? At best I can see a minor difference between the 88mb hdmi and 44mb avchd
and that takes a pretty magnified look. With the same bit rates I can't see any difference, at least on these grabs.

"A lot of breakup". You make it sound like the avchd at these bitrates is a mess.

I also don't understand why you're not doing some tests yourself. You seem to know a thing or two.

proaudio4

Thanks Ralph, that explains the noise (ISO640) in the shadows (front of house).

I happened to notice there is more visible chroma noise in the HDMI capture.
Is the chroma noise being somewhat masked by the compressed codec, or does the HDMI have
a higher contrast and increased chroma? I see the color temp difference.

Ralph_B

@cosimo_bullo

You bring up an interesting point. It makes sense that ISO would have some effect on compression since the codec interprets noise as motion.

I'm currently in the middle of doing some nighttime tests which may give us some additional insight.

cosimo_bullo

@chauncy - You're right, I'd love to be doing more real testing, but we're slammed here. That said, fairs fair so I'll at least get a few screen grabs here to show what I'm seeing. These are 640: First is the eiji 132M gop3, then next the brandin 66M gop12. Both AQ2. Both are 200% and I'm looking only at the behavior of the gradation of the light on the wall.

cosimo_bullo

Ralph, great to be talking about this. If you have any chance to do so, please do include eiji's gop 3 AQ2 in your future tests. I think he's on to something.

Ralph_B

@proaudio4

Yes, in general H264 does blur and smear out the fine noise that's in the uncompressed signal. This is not unique to the GH2. I've seen it in other cameras, as well. However, the still images are a bit deceptive in this regard. When you watch the movie playing back at speed, it appears that the HDMI has LESS noise, because it's so much finer than the compressed noise.

cosimo_bullo

And here are the same shots at 160 ISO. Again 132 first, the 66 second. Now they look much closer, and in fact are all but indistinguishable as I click back and forth. (whoops, IGNORE the last, brighter photo, it went up by accident and the site won't let me remove... mods??? Help! - please compare only the two darker photos!!!)

Stray

@cosimo_bullo I don't get why this question of it being at a Higher ISO (therefore noiser) is something worth chasing in this thread. The fact that 132M GOP 3 looks better is obviously going to be true because it's GOP3 and will therefore smear the noise less because the frequency of I-Frames is multiplied by 3. If you ran 66M at GOP 3 (which we can't at the moment) it would be indistinguishable at High ISOs to the 132M settings as it is with low ISOs (given other tests we've seen relating to extreme colour correction accross the bitrates). Image quality of a lower GOP (in terms of image noise) if you are shooting at High ISOs will be better, but thats a given we'd expect I think, it's not something that we need to test.

Edit: Granted though, if you are shooting in low light with a High ISO, want maximum IQ, and can afford the increase in file size of running at 132M (and much shorter shots) then switching to a GOP 3 is going to make sense. Though you may not like the look of the motion. Because of the way the codec works, compression is going to be worse in areas of shadow. If you shoot at a High ISO a scene with very little shadow then I think the differences between the 66M and 132M would be a lot closer accross the frame.

Stray

I reckon (probably wrongly) that the best settings accross all conditions (with reasonable bitrates) will be found with either 66M GOP 6 AQ 3 or 66M GOP 3 AQ4. The lower GOP would compensate for the drop in motion rendering quality resulting from the higher AQ as well as improving the noise in shadows (as a higher AQ will too). We just can't get those settings (yet).

Botvinic

Hello. I want to use 720 50p and 1080 50i mode! What settings should I use?

*PS What is the difference between 720 50p and 1080 50i mode (Sharpness and quality)?

Thank you!

Ralph_B

Round three of HDMI vs AVCHD

Tonights contestants are:
Chris's 44M AQ4
Chris's 66M AQ2
Chris's 66M modified to AQ3
Chris's 66M modified to AQ4
Eiji's 132M GOP3 AQ2

This is an extreme low light test. Lens: Minolta Rokkor 58mm f1.2, ISO 6400, shutter 1/50.

OBSERVATIONS:

Static -
The three 66M's and 44M AQ4 are all very close, but I feel 66M AQ3 and 44M AQ4 have a slight edge.
But the hands down winner is 132M AQ2 GOP3. It's so close to the HDMI that... well, draw your own conclusions.

Motion -
It's difficult to judge motion because there's so much motion blur at 1/50 sec shutter.
I personally feel that at high ISO's and slow shutter speeds, motion is a non-issue. All details are lost in a sea of noise and motion blur. All of the settings basically look the same as the HDMI. The only difference is how they handle the noise. And here they follow what they did in the static test. So, I'm not going to waste your time by posting the motion pictures.

Up Next -
I'm impressed with Eiji's 132M GOP3. Tomorrow I'll test it in daylight along with Chris's 66M modified to AQ3.

Chris's 44M AQ4 Static

Ralph_B

Chris's 66M AQ2 Static

Howdy, Stranger!

Categories

Tags in Topic

Top Posters