GH2 Noisefloor Index

[Sage]

Iso noisefloor index:
http://vimeo.com/18070294


Make sure to download the original file.

I think I'll be trying to light to iso 320, when without NR.
(possibly Nr for me in high iso unavoidable low light)
Interesting (perhaps revealing) NR handling of noisefloor. (don't judge noise reduction from this noisefloor test though; actual performance with lit subjects is what counts there)

UPDATE Part 2 (Lowlight stress):
http://vimeo.com/18105327
Password: part2

(ignore the diagonal pattern in this grab, that is from the screen capture, not video))

 

[rigs]

Cool! Thanks for the test.
What were your camera settings, what lens, was the cap on and iris stopped all the way down.

 

[Sage]

Cool! Thanks for the test.
What were your camera settings, what lens, was the cap on and iris stopped all the way down.

Thank you rigs!

The settings were:
Nostalgic
-2 contrast
-2 sharpening
-2 saturation
Adobe RGB

This combo really prevents the camera from crushing/compressing any noise data, by lifting it.
There was no lens, just the mount cap. Not a chance for light to enter, shutter 1/60 (what would be most common for ntsc power grid).
The export to final file used constant bitrate at 21mbps (average brate in GH2 24mbps mode) with maximum bit depth rendering to ensure that every last bit of captured noise would be transfered in full glory.

 

[Pietro Impagliazzo]

ISO 2000 with NR looks great... Is that right?

Downloading right now.

 

[Sage]

ISO 2000 with NR looks great... Is that right?

Downloading right now.

Its tough to decide on Nr + iso combos with this one, what will be really important to the final story on NR is to see a similar test, but in low light (that includes both a little tungsten and daylight).
Judging by the shots without Nr (to guess NR performance), I suspect 1600 will be a safe bet to retain rich image detail.
Thanks for checking it out Pietro! (320 without nr is kinda like a red i guess ;} )

 

[taubkin]

That was cool!

 

[Sage]

That was cool!

Thank you for checking it out.

I want to post an update, part 2 to the noise tests. (this time a lowlight stress test)
http://vimeo.com/18105327
Password: part2

Looks like strong blow out points irritate the blue channel in low light, and you'll get banding. Look out for strong light sources when shooting at night!

 

[FelixGER]

Uh Oh!
Noise is very acceptable but the banding! JESUS!

No playing with rim and backlights with this camera...useless for professional work!

 

[Sage]

Uh Oh!
Noise is very acceptable but the banding! JESUS!

No playing with rim and backlights with this camera...useless for professional work!

Not to worry. This is a stress test, and the extent of the tungsten blow out seems to cause the trouble. As long as it is lit to 320, back lights/rims should be fine (or even beyond that iso in a reasonable situation).
One of the big tricks with any cmos cam is to just get one daylight balanced CFL fill in there, and the blue channel will be happy.

 

[TheDingo]

Hi Sage,

Two questions:

1- What is your post production workflow ?

2- Why Nostalgic instead of Cinema or Smooth ?

 

[Sage]

Hi Sage,

Two questions:

1- What is your post production workflow ?

2- Why Nostalgic instead of Cinema or Smooth ?

Thanks for watching -
The original file is directly encoded in the 21mbps constant bitrate mp4 for web distro using CS5, maximum bit depth.
Nostalgic because it has the highest shadow lift, which saves the noise from the avchd compression engine. (these tend to be biased against shadows, they are based on perception psych model)
Smooth is best for natural color, I prefer to use that. Here's a nice smooth I shot today, to test 720p + 640iso:

Avoid cinema, which places midtones far from highlights in the compression.
The cam is great.

 

[Soldier]

Hi Sage. I see on your vimeo page you have some experience shooting with the t2i. How would you say the GH2 compares in terms of low light shooting?

 

[Ben_B]

That looks great, what lens did you use there just curious.

 

[Peter J. DeCrescenzo]

... One of the big tricks with any cmos cam is to just get one daylight balanced CFL fill in there, and the blue channel will be happy.

Interesting. I'm wondering (caution: stupid question ahead!), might it make sense (instead of the daylight bulb) to purposely shoot with a slightly blue filter on the lens (set WB before mounting the filter) to make the CMOS' blue channel "happy", and then color-correct the blue out of the scene in post? Just thinking out loud ...

 

[taubkin]

Interesting. I'm wondering (caution: stupid question ahead!), might it make sense (instead of the daylight bulb) to purposely shoot with a slightly blue filter on the lens (set WB before mounting the filter) to make the CMOS' blue channel "happy", and then color-correct the blue out of the scene in post? Just thinking out loud ...

We used to do that shooting with the RED on earlier firmwares, especially on green screens lit by tungsten sources.

 

[Peter J. DeCrescenzo]

We used to do that shooting with the RED on earlier firmwares, especially on green screens lit by tungsten sources.

Thanks for that acknowledgment. I'm glad my question wasn't off-the-wall after all.

Is there a particular strength blue filter you'd recommend, or does it vary depending on how a scene is lit?

Also, can this sometimes help with the weird color fringe that can sometimes be seen in highlights on skin?

 

[FelixGER]

Hmmm, I tested the noise on my GH2, too.
Mine reacts completely different! ISO 320 is worse than 160, 200, 400, 640

160 is cleanest by far

And another thing. The colortemperature changes even though the WB is set to 9000 in every shot

http://www.vimeo.com/18170823

 

[Sage]

I was traveling for a few days, I apologize for my slowness in getting back to you guys.

Hi Sage. I see on your vimeo page you have some experience shooting with the t2i. How would you say the GH2 compares in terms of low light shooting?

The gh2 is significantly better than the t2i in lowlight. Perhaps the gh2 is more comparable to the 'native' isos of the 60d and 7d.
The t2i is not the ideal dslr for lowlight shooting, and has been bested by the gh2 in just about every way (except the image quality of the lcd screen).

That looks great, what lens did you use there just curious.

Thank you Ben. That was my beloved Nikon 50mm f1.8 Series E. An old, light, full frame lens that is very organic in its artifacts.

Interesting. I'm wondering (caution: stupid question ahead!), might it make sense (instead of the daylight bulb) to purposely shoot with a slightly blue filter on the lens (set WB before mounting the filter) to make the CMOS' blue channel "happy", and then color-correct the blue out of the scene in post? Just thinking out loud ...

I see this was covered already, but I would definitely consider putting a daylight cfl in if you're already invested in tungsten, or opting for all daylight balanced. Filters will only take away light, making the red and green channels more underexposed. One cheap cfl will do wonders.

Hmmm, I tested the noise on my GH2, too.
Mine reacts completely different! ISO 320 is worse than 160, 200, 400, 640
160 is cleanest by far
And another thing. The colortemperature changes even though the WB is set to 9000 in every shot
http://www.vimeo.com/18170823

I wanted to see your test, but it was not available. From what you are saying, it sounds quite similar to the results of my low light test (part 2).

I replicated my low light experiment in two different conditions after that initial test, with tungsten and daylight balanced, and the results seemed to hold. The best isos continuously proved to be (with notes):

160 - this I have concluded is the best iso, the combination of best accuracy with low noise make it stand out (Telecrop feature shows the 1:1 pixel level, and 160 iso is the only one you will ever want to use to get clean images in telecrop)
320 - without NR, this iso is quite clean, not far from 160 - but it is inaccurate in the shadows; it seems to suppress the blue channel in the shadows to achieve its clean quality; in actual use this does have a nice filmic feel to it
400 - this is more accurate in the shadows, but at the cost of blue noise... the noise signature is a bit more aggressive than 320, and the noisefloor (at black) has noise while the 320 crushes it into smoothness
640 - what I will call the last "clean" iso... noise enters the midtones a bit, and is more aggressive in the shadows, but the image stays accurate and is not yet 'obscured' by a layer of noise
1250 - noisy, but the best deal overall of the noisy isos

Notes on 200, 400, 800, 1600
- each of these takes on a slight to moderate 'layer' of blue noise compared to the iso immediately below.
200 isn't bad, but I think to myself, if you have that much light - why not go for the king 160! Its worth it.
400 as noted is more accurate in shadow, so if you needed that you could use neat video and have a deal, though 320 has a nice 'grainlike' film feel to its shadows, and is better than 400 when not processing for noise
800 takes on the blue noise layer, but its more aggressive than the earlier isos - I think, why not push to use 640 as a medium iso, it is a big 'switch point' in terms of level of noise
1600 this would still be a relative deal compared to going up to 2000, but hopefully you have enough light to opt for the better value of 1250. You'll have to hit either with Neat Video.

 

[FelixGER]

Hi,
I repeated the test and deleted the first one because it wasn´t very good
http://www.vimeo.com/18240143 (Download)

In my opinion 400 is better than 320 und 640 is really bad and I choose 800 instead.
What do you thing?
But it is weird that when changing from 640 to 800 the color temp. changes also

 

[Sage]

Hi,
I repeated the test and deleted the first one because it wasn´t very good
http://www.vimeo.com/18240143 (Download)

In my opinion 400 is better than 320 und 640 is really bad and I choose 800 instead.
What do you thing?
But it is weird that when changing from 640 to 800 the color temp. changes also

Yes the color rendering definitely seems to change with iso. And I see what you mean in your selections... Unless the temperature of light you are using is producing a different effect (a possibility) - it looks like there may be a difference in manufacturing continuity.
I have reproduced my results several times, yet your single test puts a different spin on what I've seen. What I would now like to see is this same test replicated on different gh2s - to see if it is in fact varying by manufactured chip.

-Also, I noticed there was no ISO 250 - did you intentionally skip that? I ask because your 320 looks like my 250, and the 400 looks like my 320.