PDA

View Full Version : Noise/ISO comparison



Barry_Green
11-11-2009, 09:01 PM
After much talk about the low light advantages that DSLRs offer, and a recent discussion about noise versus sensitivity, I decided to see how the wide ISO range of the 7D and GH1 compare, as for overall sensitivity and noise performance, against a conventional video camera with reasonably good low light performance (in this case, the HPX170).

I tweaked them all to give what I figured would be good noise performance; I took the sharpness down on all of them, I used the 320/640/1250/2500/5000 ISO's on the Canon, I boosted the coring up and used B.Press on the HPX170, and I used the GH1 in fairly neutral settings, 320/640/1250 ISO, with sharpness down.

Then I shot a chart (oh yes, a chart again!) at various ISOs, changing the shutter speed to keep the overall exposure the same (so, 320 ISO @ 1/60, 640 ISO @ 1/125, 1250 ISO @ 1/250, 2500 ISO @ 1/500, and 5000 ISO @ 1/1000, and 6400 ISO @ 1/1250). I did the same thing with the video camera, using 0dB of gain @ 1/60, 6dB @ 1/120, 12dB @ 1/250, and 18dB @ 1/500.

Which means, basically, if I translated the video camera's base ISO plus gain, it could be said that I was testing the video camera at 500, 1000, 2000, and 4000 ISO.

The DSLRs employ some pretty aggressive noise reduction, especially the GH1, and it's effective. At 320 the noise is pretty minimal, and even at 640 the noise is a less than the video camera at zero dB. I think the GH1 at 1250 ISO compares favorably, strictly on a noise-performance basis, with the video camera at 500 ISO.

From that point on, though, they pretty much trade places, and are much closer in noise performance than I ever would have expected. When you compare noise for noise, ISO against gain, they look pretty evenly matched. Far more evenly matched than I would have thought.

So -- judge for yourself, I've arranged them in what I consider ascending order of noise performance. Test subject was the grayscale portion of the Billups VF/X 1+ chart from DSC Labs, cropped and extracted and blown up to 200% size for easier exhibition of the noise.

Note: this is not a "low light" test, this is "how does the sensitivity compare, along with the grain?" In some ways you could look at it as a low-light test, but both DSLRs were at f/5, the video camera was at about f/4. The SLRs could open up another half-stop, but the video camera could open up another two and a half stops. Then again, the SLRs could have f/1.4 lenses on them, if we don't want to compare stock lenses, so ... no, this isn't a "low light" test. This is just to see how the noise and sensitivity compare, to see if there's some big advantage in the new DSLR technology.

Be sure to watch it full-screen, to exaggerate the noise patterns and see the differences.

7a96d5b51511e1c5f5

Ben_B
11-11-2009, 09:07 PM
What was the noise reduction setting on the GH1 at? You said you reduced sharpness, what about NR? I usually shoot -2 (more noise.)

Barry_Green
11-11-2009, 09:09 PM
I actually had it at zero. I could have gone to +2, which would have been more fair with the HPX170's coring +7 setting. But in general use I normally run it at -2, so I kind of just punted and went with zero. It's really clean anyway. At least as far as crawling noise goes.

There's a second part to the test involving the CamBelles and the picture turning to oatmeal, but it's a lot harder to quantify (because downrezzing makes everything look better, so you're not seeing what's really happening).

Ben_B
11-11-2009, 09:12 PM
Good deal, thanks Barry.

Kellar42
11-11-2009, 10:07 PM
Barry thanks for more tests, just one question. I know this isn't a 'low-light' test, but in what other situation would these noise sensitivities crop up? (The reason I'm asking is, okay the video camera may do better at extremely high ISOs, but wouldn't the DSLR wide open be able to go into darker situations with less ISO in the first place?) Is there somewhere else where the way these cameras handle and produce noise becomes an issue?

Barry_Green
11-11-2009, 10:22 PM
Well, like everything, "it depends". The video camera, zoomed out to wide angle, opens up to 1.6. You're going to be pretty hard-pressed to find an f/1.6 wide-angle lens that matches what the video camera can do. But, then again, the relative sensitivity of the video camera isn't necessarily a linear performance curve...

At an equiv. of 50mm, the video camera is going to have a max opening of, oh, maybe around f/2.0. The DSLRs, with their stock lens, are going to be at f/4 or f/5. But you could replace the lens with a wide-open f/1.4 type of lens and get a lot better performance.

I guess all that's going on here is that the DSLR is claiming ISO 1600 or ISO 5000, and the video camera just has "gain", but -- when it all comes down to it, they're not that different. They all perform about the same, noise/sensitivity-wise.

The GH1 doesn't get grainy/noisy like the others, but its image goes soft and the color saturation goes down the tubes. I'd stick with 800 or lower on the GH1 whenever possible.

I guess I was just really surprised; we always complain horribly about "gain" on video cameras, and people trumpet the "great" low-light capabilities of the DSLRs, but when you get right down to it -- they're about the same.

philiplipetz
11-12-2009, 03:32 AM
Barry, thanks. I am travelling and only have my iPhone so I cannot view the video, but you indicate two different patterns of degradation so which of them is better rescued in post using normal tools, and using a noise reducer plugin (without getting a plastc look). GH1 or the pack of other cams?

philiplipetz
11-12-2009, 03:37 AM
Barry, a second question. In your opinion how much of the perceived low light dSLR video advantage is due to aliasing covering up noise? And if that is what is happening will the viewer perceive an advantage not visible in charts? Does this give the 7d with it's greater aliasing a perceived advantage over the GH1?

Martti Ekstrand
11-12-2009, 03:53 AM
When I've somewhat sloppily mentioned that HDSLRs 'are better in low-light' I've always meant with a 35mm adapter on the video camera for shallow DOF images. How does the 170 + 35mm adapter fare in a test like this?

philiplipetz
11-12-2009, 03:55 AM
even if cams are technically smilar there is a big price difference.

maarek
11-12-2009, 04:29 AM
And I have to disagree. My low light tests with 6400 ISO look way better.

http://www.vimeo.com/7128026

Though I don't shoot test charts. Also any videocamera I've tried doesn't come as close. Not even the RED 1.

Chris_TC
11-12-2009, 05:59 AM
I guess I was just really surprised; we always complain horribly about "gain" on video cameras, and people trumpet the "great" low-light capabilities of the DSLRs, but when you get right down to it -- they're about the same.
I appreciate your tests, even though in this one the compression was so bad that I didn't see any noise at all, just big blotches. I had to assume you had put the shots in the right order because based on the web footage I wouldn't have been able to.

We all know that small sensors produce more noise. And yet there are compact photo cameras that shoot essentially noise-free images. How can this be? Well, because they noise reduce the heck out of the images. And it shows, because the resulting photos look like watercolor paintings.

Would you notice the watercolor look on a test chart with a bunch of solid colors? No, you wouldn't.

I don't see how a small sensor could possibly keep up with the noise performance of the larger sensors unless it uses significant noise reduction. Isn't this what physics dictate?

Kellar42
11-12-2009, 07:40 AM
But Barry, wouldn't a video camera and DSLR, both at F1.4, both at the same ISO have different low light performances due to the size of the sensor?

Ian-T
11-12-2009, 07:46 AM
But Barry, wouldn't a video camera and DSLR, both at F1.4, both at the same ISO have different low light performances due to the size of the sensor? Maybe that’s the point he’s making….that it's not necesarily just the sensor size and that if these video cameras had the ability to exchange lens like these DSLRs then the differences in low-light ability would not be that far apart (or maybe I’m way off).

Kellar42
11-12-2009, 08:09 AM
Maybe thatís the point heís makingÖ.that it's not necesarily just the sensor size and that if these video cameras had the ability to exchange lens like these DSLRs then the differences in low-light ability would not be that far apart (or maybe Iím way off).


Well, most of the video cameras go pretty open as far as their apertures, so I'm not sure the lenses would make that much of a difference. It sounds like he's saying there isn't much of a difference period, which is what I'm trying to verify...

Barry_Green
11-12-2009, 10:09 AM
Barry, a second question. In your opinion how much of the perceived low light dSLR video advantage is due to aliasing covering up noise?
None. Noise is inherent in the sensor, and will appear over everything.

Barry_Green
11-12-2009, 10:10 AM
When I've somewhat sloppily mentioned that HDSLRs 'are better in low-light' I've always meant with a 35mm adapter on the video camera for shallow DOF images. How does the 170 + 35mm adapter fare in a test like this?
Oh, I'm certain that in that case, the DSLR would slaughter. Most adapter require the video camera to zoom in, thus losing some of their wide-open advantage, and then there's at least a half stop of light loss from the adapter, and then cumulative light loss from the lens on the front.

For shallow DOF purposes, no question, the HDSLRs rule the roost.

Barry_Green
11-12-2009, 10:11 AM
even if cams are technically smilar there is a big price difference.
Unquestionably.

I'm not trying to "prove" the "superiority" of the video camera here. I'm just trying to bring perspective to the situation, is all -- how much of an improvement is there, if any, and if there is any, WHY is there any? That's what I'm interested in.

Barry_Green
11-12-2009, 10:16 AM
I appreciate your tests, even though in this one the compression was so bad that I didn't see any noise at all, just big blotches.
No, the compression doesn't affect it at all. That's why I did it the way I did it -- I blew up the scene to4x size to exaggerate the size of the noise. What you're calling blotches is the actual noise, just blown up 4x. I extracted an SD-sized patch out of the HD frame, then blew that up 4x.

Watch it in full-screen. Anything you see that is moving at all in any way, is noise.


We all know that small sensors produce more noise.
Small pixels do, yes. Not necessarily small sensors.


And yet there are compact photo cameras that shoot essentially noise-free images. How can this be? Well, because they noise reduce the heck out of the images. And it shows, because the resulting photos look like watercolor paintings.
Yep -- there is huge, massive amounts of noise reduction going on in the SLRs, primarily in the GH1 (and that was with NR set to middle, not even +2).


I don't see how a small sensor could possibly keep up with the noise performance of the larger sensors unless it uses significant noise reduction. Isn't this what physics dictate?
If the pixel count was the same, I'd agree with you. But in this case, the actual pixel size is pretty much the same between the two sensors. The HPX170's pixel size is about 5.5 square microns, the 7D's pixel size is about 4 microns. Then you have to factor in CCD vs CMOS, and active noise reduction vs. not so much, and spatial offset vs. pixel binning...

Barry_Green
11-12-2009, 10:17 AM
But Barry, wouldn't a video camera and DSLR, both at F1.4, both at the same ISO have different low light performances due to the size of the sensor?
Why would they? ISO = ISO, whether you're talking about a 1/6" video camera or a frame of 35mm film, if the rated sensitivity is the same, the sensitivity should be the same.

The noise/grain may differ, but the sensitivity will be the same (and that's really what I wanted to test here, was noise at sensitivity levels).

ydgmdlu
11-12-2009, 11:14 AM
If the pixel count was the same, I'd agree with you. But in this case, the actual pixel size is pretty much the same between the two sensors. The HPX170's pixel size is about 5.5 square microns, the 7D's pixel size is about 4 microns. Then you have to factor in CCD vs CMOS, and active noise reduction vs. not so much, and spatial offset vs. pixel binning...
This is the first test that I've seen that actually provides evidence for sensor size, in itself, being only a minor factor, at best, in sensitivity and noise performance.

Barry_Green
11-12-2009, 11:45 AM
I know it's not the test folks want. They want a "low light" comparison; I'm just not sure how to do that and have it mean anything.

For the record, I did just put the 7D on 6400 ISO and the HPX170 on 18dB and stick 'em in my closet, and the 7D's picture was quite a bit brighter, and noisier (which you'd expect, since the 170's ISO equiv. is 4000 vs. 6400), so -- how do you draw the line? How do you decide what's fair?

In any case, all I was really looking for is where the "magic" is, and I don't think there is any, a sensor is a sensor and performs based on native sensitivity (ISO) and gain. Noise reduction can improve the usable sensitivity range; if there's any magic anywhere, that's where I suspect it is.

Chris_TC
11-12-2009, 02:44 PM
If the pixel count was the same, I'd agree with you. But in this case, the actual pixel size is pretty much the same between the two sensors. The HPX170's pixel size is about 5.5 square microns, the 7D's pixel size is about 4 microns.
But doesn't the pixel binning effectively increase the pixel size of the SLRs or is that not how it works?

ydgmdlu
11-12-2009, 03:06 PM
But doesn't the pixel binning effectively increase the pixel size of the SLRs or is that not how it works?
Barry's test shows that it doesn't help much at all.

Barry_Green
11-12-2009, 03:28 PM
But doesn't the pixel binning effectively increase the pixel size of the SLRs or is that not how it works?
Yeah, that's a puzzler. Canon says they don't skip rows, they bin pixels. Binning has traditionally been used as a way to increase pixel size, with the added benefits that you get better relative sensitivity and lower noise. But the footage shows none of those benefits, so -- I don't know what the answer is. It's tempting to go back to the notion that they really are just skipping rows. But they directly denied it, so ???

Ian-T
11-12-2009, 07:19 PM
Yeah, that's a puzzler. Canon says they don't skip rows, they bin pixels. Binning has traditionally been used as a way to increase pixel size, with the added benefits that you get better relative sensitivity and lower noise. But the footage shows none of those benefits, so -- I don't know what the answer is. It's tempting to go back to the notion that they really are just skipping rows. But they directly denied it, so ???One piece of evidence might be the hit in resolution....right? I mean bigger pixels should mean better light sensitivity yes...but at the same time a hit in resolution....no? If yes then would that explain the cameras resolving issues on the chart? Just guessing here.

Jason Ramsey
11-12-2009, 07:21 PM
there would be a hit in the resolution from skipping lines as well...

later,
jason

stephenvv
11-12-2009, 10:39 PM
Barry,

I realize you a very knowlegable guy but this test seems of questionable value. First, I can't accept (and it's visually very apparent) that re-compression and web compression took a big toll on these images.

Noise is entirely a contextual issue - size, quality, dynamic range, resolution all are factors in judging if noise (as well as film grain) is an good/bad. Thus it's a subjective issue and only valid if we are looking at camera original images.

Second - the GH1 and 7D have been extensively tested for noise online in far more comprehensive, careful and scientific manners. The 7D wins hands down especially at 800 plus - so your comment "they are about the same" seems odd.

Finally, I've shot with a lot of camera and despite having just got a 7D still at factory defaults, it's low light shooting is breathtaking compared to other cameras - specifically the lack of noise.

Of course I'm judging this HDMI at 1080p. But I just had my first paying gig at a haunted attraction. I was shooting in absolute darkness, no moon, way in the country.

Canon 50mm f/1.4, ISO 1600. Here's a straight PNG from Vegas off camera original and one with a bit of NR and levels.

http://www.sv2studios.com/media/1600.png


http://www.sv2studios.com/media/1600-2.png

Barry_Green
11-13-2009, 04:14 AM
I can't accept (and it's visually very apparent) that re-compression and web compression took a big toll on these images.
Remember they're zoomed in 4x. And no, web re-compression took basically no toll, that's what the noise patterns look like. I could post the original file if you want.[/quote]


Second - the GH1 and 7D have been extensively tested for noise online in far more comprehensive, careful and scientific manners. The 7D wins hands down especially at 800 plus - so your comment "they are about the same" seems odd.
I'm showing you the results I got. The noise looks higher to me on the 7D. I can see noise on the LCD at 800 ISO, whereas on the GH1 it doesn't appear to get noisy, it just appears to get mushy and the color saps out of it. The 7D's image doesn't suffer from those effects, it stays consistent in color and sharpness, it just gets noisier.

The GH1 appears to me to be using highly aggressive noise reduction, and as a result at 1600 ISO the image goes soft and muted, not unlike the HMC40 does. The 7D doesn't seem that way to me, it looks more like what a conventional video camera does -- the more you gain it up, the noisier it gets. But it gets very chroma-noisy, very very much like the older HVX200, when you get really high up there (5000 & 6400).


Finally, I've shot with a lot of camera and despite having just got a 7D still at factory defaults, it's low light shooting is breathtaking compared to other cameras - specifically the lack of noise.
At high ISOs? Are you saying that the noise results you get are unlike mine? I would find that a huge relief and great news, but I doubt it's true, I think you're not looking for it. Isolate it out like I did, and I bet you a nickel that your has exactly the same noise that mine does (although I would love love love to be proven wrong on that). Do you have any sort of gray chart you can shoot and test that with?

Simple test -- put it on 6400 and use the 10x magnified viewfinder. Point it at some wood. On mine there's a snowstorm of noise, no need to even record it or blow it up, it's there and it looks noisier than a +18dB HPX170, no question.

Same thing with 1600 ISO -- use 10x, point it at your desk (mine is a sort of light cherry orange-ish color) and I'm seeing a screen that's crawling with noise. The blacks are pure black, but bring it gradually up and it gets obviously quite noisy. You can even do it with a blank sheet of paper, stop the iris down until it's closed, then open a little bit and a little bit more. If you say that yours is noiseless at x10 doing that test at 1600 ISO, then there's definitely something wrong with mine. If, however, yours is showing the same crawling noise pattern that mine does, then I assert there's nothing wrong with this test at all, and it goes to show what should basically be pretty obvious anyway -- sensors with roughly equivalent pixel size deliver roughly equivalent noise. (unless massive noise reduction is employed, as the GH1 appears to do, which leads to different artifacts as mentioned before).

And stills aren't the way to show noise, because noise moves. Blow it up and isolate it.

Barry_Green
11-13-2009, 04:27 AM
Okay, here's a camera-original raw file. No recompression, no web compression, this is straight off the CF card. ISO 6400, pointed at my desk, out of focus, at about f/4, not dark but I kept stepping through the iris until I found one that really shows off the noise. This is my attempt to isolate the noise to show the pattern I get, in case you think it's abnormal or unusual.

Scene file Standard, with sharpness turned down to zero, contrast up one notch, and color saturation down two notches.

http://www.mediafire.com/download.php?ygzjxizzugv


If you could do something similar and compare it to what I'm getting that'd be appreciated.

Kellar42
11-13-2009, 07:46 AM
Second - the GH1 and 7D have been extensively tested for noise online in far more comprehensive, careful and scientific manners. The 7D wins hands down especially at 800 plus - so your comment "they are about the same" seems odd.


Have they, and against each other? In the video mode? Any links?

Also Barry, thank you, I didn't realize that ISO was ISO across the board and that ISO on one of these sensors is supposed to correspond with the same sensitivity film stock. The people out there getting different readings of their light meters than what the camera says, etc, are what contributed to my idea that there are differences.

stephenvv
11-13-2009, 11:12 AM
DP Review just posted their in-depth on the 7D and I quote "Despite the highest nominal resolution of all APS-C DSLRs and therefore a very small pixel-pitch the EOS 7D performs very well in low light situations and manages to maintain a good balance between image detail and noise reduction up to very high sensitivities. It's visibly better than the EOS 50D and as good as it gets in the APS-C class (if you prefer the 7D or Nikon D300S in this respect is probably a matter of taste). If you require significantly better high ISO performance than the EOS 7D can provide, your only option is to move into the full-frame segment."

6400 ISO is quite noisy on a 7D and unusable on a GH1 but in my total darkness gig, complete unneeded. If my eye could see, the 7D and 1.4 lens gave me shots at 800 (which is very clean if exposed properly) and 1600. I only used 3200 if there was minimal light (e.g. once are lit by a 20 Watt Blacklight that was nearly dark.

The factory setup in the 7D favors slight underexposure to keep detail in highlights. If you are seeing too much noise, make sure you are not overexposing or use a different picture style.

Imaging Resource has direct comparisons with 7D and GH1, no contest at 800/1600/3200 especially shadow noise, detail and dynamic range.

I've not seen a proper video test but from taking stills while shooting footage, I see no difference in anything - noise, dynamic range etc. still to video.

Until somebody does a careful controlled test of noise like the above sources, I think the still results are the best info presented yet.

Although I can't back it up scientifically, it's clear to me the 7D is fantastic low light camera compared to anything I've used before. I wouldn't mind having a Mark IV as well when budget allows but the 7D is great camera, still or video.

Barry_Green
11-13-2009, 11:23 AM
Well, ISO stands for International Standards Organization, so -- it's *supposed* to be standardized! :)

Whether it actually is, or not, is obviously something open to interpretation. I found them relatively comparable across the board.

Barry_Green
11-13-2009, 11:26 AM
DP Review just posted their in-depth on the 7D and I quote "Despite the highest nominal resolution of all APS-C DSLRs and therefore a very small pixel-pitch the EOS 7D performs very well in low light situations and manages to maintain a good balance between image detail and noise reduction up to very high sensitivities. It's visibly better than the EOS 50D and as good as it gets in the APS-C class (if you prefer the 7D or Nikon D300S in this respect is probably a matter of taste). If you require significantly better high ISO performance than the EOS 7D can provide, your only option is to move into the full-frame segment."
Yes, but that's about stills, and the 7D has the high noise reduction on stills capability.


6400 ISO is quite noisy on a 7D and unusable on a GH1 but in my total darkness gig, complete unneeded.
Well that's what I was asking you -- is my 6400 the same as yours? And it's unusable on the GH1 because it's not even offered, the GH1 tops out at 1600. You can't go higher than 1600 on a GH1 in video mode.


If my eye could see, the 7D and 1.4 lens gave me shots at 800 (which is very clean if exposed properly) and 1600.

[quote]Imaging Resource has direct comparisons with 7D and GH1, no contest at 800/1600/3200 especially shadow noise, detail and dynamic range.
Which sounds like it's still performance, where noise reduction comes in. From my testing it looks to me like the GH1 employs aggressive noise reduction in video, and the 7D doesn't.


Until somebody does a careful controlled test of noise like the above sources, I think the still results are the best info presented yet.
Well, my question to you is -- how is what I did not a careful controlled test of noise? What did I do wrong?

stephenvv
11-13-2009, 11:32 AM
Here's some screen captures that may help clarify from Imaging Resource's comparometer. At ISO 800 the GH1 appears to be doing well with noise but if you look at the full range of shots, you will see it's far more aggressive - thus leading to detail loss AND the more limited dynamic range of the GH1 (again as tested by several sites) is clearly visible as loss of detail - crushed blacks and white clips compared to the 7D. But if you look around the ISO 800 GH1 images, you will see evidence that noise is higher and is being aggressively processed out.

But at ISO 1600 and 3200, the GH1 NR cannot keep up and the 7D is way ahead. Only a 5D, Mark IV or D3s etc. will fare better than a 7D.

ISO 800

http://www.sv2studios.com/media/7d-gh1-800.png

ISO 1600

http://www.sv2studios.com/media/7d-gh1-1600.png

ISO 3200

http://www.sv2studios.com/media/7d-gh1-3200.png

stephenvv
11-13-2009, 11:43 AM
[/i]
Well, my question to you is -- how is what I did not a careful controlled test of noise? What did I do wrong?

I'm not sure I agree that the 7D and GH1 alter NR just for video - is that definitively known? As far as my eye can tell, the 7D does the exact same NR on stills and video.

As to why I don't think this is a very scientific test --

If you look at the stills ISO noise testing from leading places like DP review, Imaging Resource, they are measuring (typically with software) the amount of noise, the amount of NR (detail loss) and the amount of resolution retained. They rarely shoot charts as cameras are designed to shoot real things, so they typically set up a controlled real scene or simulation.

Then these results are charted against dynamic range - poor dynamic range can hide noise by crushing blacks and blowing out whites, so you need a total picture of sensor performance when used in real world situations.

I've seen that the results reported on those sites are an excellent indicator of how cameras will behave in demanding shooting conditions. That's what we all really want to know.

So it's not the the 7D is "noise-free", it's that the total design of sensitivity, noise, noise reduction, dynamic range lead to a camera that handles low-light shots better than any APS-C camera (or obviously any smaller sensor camera). That jibes with my real world experience with the 7D thus far.

Barry_Green
11-13-2009, 01:12 PM
Okay, but none of that was under discussion. DP Review and others are testing stills. I'm looking for the noise performance on video/moving images. Nothing about stills performance is relative here, and yes I've verified that setting the "high ISO Noise Reduction" setting to "strong" has no effect on the video.

And again, I wasn't doing a "low light" test. I was looking to see what the noise performance was, between various cameras and technologies at various ISOs. Low light means lots of things, including lens usage and accounting for the nonlinear light performance of a videocamera like the HPX170 when in the most-open f-stops, etc.

I reject the idea that folks seem to sometimes say that "charts don't reflect the real world". Charts specifically show exactly what is happening, in the real world and in the lab world. That's what the charts are designed to do -- if they didn't, they'd be pretty pointless and shouldn't be costing $400 to $1500 each! :D Furthermore, I really don't think the camera has some sort of "chart intuition" where it says "hmm, I'm shooting a test chart, I'd better perform horribly, but as soon as I shoot a "real world" scene then I can go back to performing well." The performance is exactly the same in both scenarios, the charts just let us isolate and understand what the unit is actually doing.

So the chart lets us know what kind of noise exists, and where it manifests itself. Knowing where the noise exists lets you know, for example, how to light around it. If the noise shows up in a band of 10 to 30 IRE, like on the HVX, then a wise shooter could try to light his scenes to avoid having too much image in that band. Darken the 10-ire stuff, and brighten the 20-ire stuff, crush it down out of or lift it up out of the noise band, and voila -- much cleaner footage.

Unrelated to all that, can I ask again --did you look at the 6400 iso clip I sent, and does the noise in my clip line up with the noise in your clips? Is there a difference in how our 7Ds are performing?

stephenvv
11-13-2009, 01:32 PM
I'm looking for the noise performance on video/moving images. Nothing about stills performance is relative here, and yes I've verified that setting the "high ISO Noise Reduction" setting to "strong" has no effect on the video.

"high ISO NR" on which cam, the GH1 or 7D?


And again, I wasn't doing a "low light" test. I was looking to see what the noise performance was, between various cameras and technologies at various ISOs. Low light means lots of things, including lens usage and accounting for the nonlinear light performance of a videocamera like the HPX170 when in the most-open f-stops, etc.

I reject the idea that folks seem to sometimes say that "charts don't reflect the real world". Charts specifically show exactly what is happening, in the real world and in the lab world. That's what the charts are designed to do -- if they didn't, they'd be pretty pointless and shouldn't be costing $400 to $1500 each! :D

I'm not rejecting charts - but I think for evaluating how noise interferes IQ you must shoot real world setups to make the judgment. Charts are flat, reflective items with qualities often not present in what is shot in the real world. I look at charts as diagnostic instruments not qualitative ones.



Unrelated to all that, can I ask again --did you look at the 6400 iso clip I sent, and does the noise in my clip line up with the noise in your clips? Is there a difference in how our 7Ds are performing?

It took several tries as the site was down but I did look at it. But without recreating your setup precisely, again, it does not tell us much. It's looks 6400 or 12500 ISO footage.

6400 ISO is only useful for emergencies on the 7D IMO - 3200 ISO can be okay in some situations where blacks can be crushed and/or end result is lower rez (web, DVD, SD). 1600 is often usable if exposed well at 720 or 1080 and nearly always good for lower rez. 800 and 640 can be used in almost all situations for all outputs.

That does not apply in my experience or the experience/footage of most other GH1 shooters - 800 and esp. 1600 are too noisy with banding issues at full rez with GH1 - but the 7D seem to be the complete opposite - only occasionally problematic at 800 to 1600.

That's why your conclusion "they are about the same" is something I took issue with.

Barry_Green
11-13-2009, 01:49 PM
"high ISO NR" on which cam, the GH1 or 7D?
7D. The GH1 has variable noise reduction settings, from -2 to +2, but with the 7D I haven't found any noise-reduction control other than C.Fn II, which again appears to only affect stills.



It took several tries as the site was down but I did look at it. But without recreating your setup precisely, again, it does not tell us much. It's looks 6400 or 12500 ISO footage.
Well, that's all I was asking, is if mine was performing on par with yours.

Maybe tonight I'll put the 170 on +18 and the 7D on 4000 (which, according to my charts, should be about roughly equivalent in ISO and noise) and try to shoot the outside of the house or something, on video, and see what the differences are.

stephenvv
11-13-2009, 02:05 PM
That makes sense - I don't have High ISO turned on on my 7D, so video and stills look the same.

If you do shoot outside, a range at the same f-stop from ISO 100 up would be very interesting instead just 4000 or +18db which I think is too noisy on both those cams.

I tend to underexpose slightly instead of boosting ISO/gain and bring up in post if I don't have enough light.

Barry_Green
11-13-2009, 02:12 PM
The problem with a range of f-stops is that the video camera actually appears to change its relative ISO depending on how much light is available to it (meaning a nonlinear response to brightness). So at f/2.8 with no gain it might be ISO 400, at f/8 it might be ISO 500, and at f/11 it might be ISO 800...

Interesting that you choose to underexpose and bring it up, as I would expect that to magnify noise; I typically advise folks that if they want a clean dark image, overexpose and then bring it down. But of course that presupposes that they have enough light to over-expose! :D

But yes, I'd rather underexpose than gain up in camera, simply because it seems reasonable that post-boosting could be cleaner than in-camera boosting (but then again, that's something that should be tested, because post-boosting would take place after h.264 compression, whereas in-camera boosting works on the unprocessed raw signal... no end of things to test...)

stephenvv
11-13-2009, 02:26 PM
That's strange about camera changing relative ISO with f-stop. I've not seen that with most other cams though it does seem that the HV20/30/40 are doing weird stuff having just wrapped a zero budget feature with them and often not having enough light.

Still, it even if it shows weird stuff, it would be interesting to see.

The reason of underexposing when I don't have enough light i.e. I'm not "trying" to underexpose, it's just then when there is not enough light for correct exposure is instead of adding is that by keeping the ISO as low as possible and close to ideal ISO, it means less noise period vs. boosting ISO.

However, with most digital cams I do try to expose to the right of histogram to keep noise down when I do have enough or close to enough light as slightly overexposure dropped in post does nicely push noise down.

The big difference of an HV30 vs 7D is with HV30, I often did not have enough light. With 7D, I often have too much.

Believe it or not, I overexposed at 3200 and 6400 using a $20 15 watt LED flashlight out to about 10 feet with the 7D. I did better turning the light off dropping down to 800 ISO. This was shooting in complete darkness. Even at ISO 800, at wide open at F/1.4, I had overexposure issues in highlights if too close to subject. And 800 is pretty clean with pleasing subtle noise texture on the 7D - I love the look.