"clean" isos and focal length applications
 

Hi guys. Two questions I've been thinking about, and haven't gotten much from searching on here or googling.

So, first, I've come across this claim several times that on the 5D, the "real" iso numbers are the ones that are multiples of 160. . .so 160, 320, 640, 1250, etc. The numbers in between those are allegedly interpolated from the "real" numbers and are therefore prone to more noise or something. The 160 multiples are supposedly cleaner. Anyone have any insight into whether this is BS?

Another question I had involves lens focal lengths as the apply to common usage in the world of TV/film. I've never worked with a 35mm sensored cam before, and I'm about to shoot a short film with the 5D.

We have several lenses available (thanks to a generous girlfriend): 24-105 zoom, 50mm prime, and a 75-300 zoom.

I'm basically curious as to what are common focal lengths for different types of shots. Obviously some of this is up to artistry and one can shoot CUs with a wide lens for several reasons, or compress a WS with a telephoto lens some reasnon, but I'm assuming there are more typical ways a 50mm lens is used vs a 24mm lens. Can anyone enlighten me? Thanks.

That whole ISO thing was based on people who didn't know what they were doing. The native ISOs are the ones you would expect (i.e., 100, 200, 400, etc.).

Well, the answer is different when you're talking about timelapse video (raw stills) vs. regular video (h.264).

For raw stills, the 200, 400, 800, 1600 ISO settings are "real" in that they do not have any in-camera multiplication nonsense applied to the raw files. 160, 320, etc. are "fake" in the sense that they start out the same as the "real" ISO settings, but Canon applies a simple digital "pull" to them 1/3 stop darker. In a well-designed system this would result in slight quantization error. But fortunately (or unfortunately, depending on how you look at it), Canon's system is not well-designed, so the extra 2-3 bits that normally go to waste serve to hide this nonsense. For the other "fake" ISO settings (125, 250, etc.), Canon increases brightness 1/3 stop with digital gain. This has two results: first it would again result in slight quantization error if Canon had tighter tolerances, but again is cancelled out by their mistakes in other areas. Second, it reduces dynamic range by 1/3 stop, as it is the highlights that get pushed off the edge in the linear digital push.

Now, for video, everything above still occurs, but we now have to consider the raw conversion and post processing that happens in-camera. For the "real" ISO settings (200, 400, etc.), Canon inexplicably sets the white point of the JPEG at 1/3 stop less than the actual white point of the raw files. This means 1/3 stop of highlights are clipped for no reason. For the "fake" ISO settings (160, 320), Canon leaves the white point the same (as if it was still 200, 400), so that 1/3 stop comes into usage. EDIT: Since the raw data is pulled down 1/3 stop, this equals out to the same highlight headroom, the net result is 1/3 stop cleaner shadows.

In summary: for raw stills, 200, 400, 800, etc. are just as good as 160, 320, 640, etc. For video, 160, 320, 640, etc. are not braindamaged like 200, 400, 800, etc., so they are (slightly) better.

Hope that helps.

How intriguing. Thanks.

I use 160, 320, 640, etcetera... religiously.

If you took a clip at 320 and then took a clip at 400 and showed me both back to back, I bet I wouldn't be able to tell.

For what it's worth.

They did some testing during those amazing Zacuto episodes and concluded the same thing, that 160 and the multiples of it are the best for video.

Sometimes things that don't show up on a computer screen/monitor/TV manifest themselves quite nastily on larger screens, say at a festival screening. So if you guys say it's so, I'll take it as doctrine.

I've watched our short from last year, shot on a 5D on a 40ft screen. I learned about this 160, 320 thing in the middle of shooting. So we used both.

I'll tell you this. If you light your scenes properly, you'll have a DEVIL of a time picking up noise even on a large screen. We shot up to 640 ISO, but I'd say that anything below 400 is the safe zone for the big screen.

I'll stick with the multiples thing. Unfortunately we have a few outdoor night/street shots where I'd like to see the backgrounds, so I may go up to (gasp) 1250 or (gasp harder) 1600 on those with the 1.8 prime. If we get noise we get noise.

Believe me, I did a shot for that movie outdoors about 8pm. Sunset was about 5:45. We shot that at ISO 640 and it was noise free, and the background was fine. Go shoot some tests in similar conditions from 320 up to 1600 and see what you think. Blow it up 400% and see what you think, because that's what its going to be like for people sitting in the front of the theater.

Well, for the night stuff, I think the ISO chooses me rather than the other way around, right? If I'm already open to 1.8 and I need it bright enough to see the backgrounds, I need to raise the ISO to whatever it takes to get there. Otherwise I would think the idea would be to shoot as low as you can, right?

Yea, that's pretty much it. But I'd experiment with adding that last stop in post, versus adding it in the camera on production day. Do some tests and see which you like better.

Sorry I'm unclear. . .do you mean going no higher than 800 and gaining up in FCP vs shooting at 1250/1600?

Yes, that is precisely what I mean.

I did a little test. 800 doesn't look half bad, actually. At least on my Panasonic/Viera Plasma. . .played through the cam via HDMI. I don't know if that's a good way to judge.

Still curious about the focal length thing.

Duh. Just realized the next "real" iso down from 1250 is 640, not 800. But you guys are saying what makes these "real" isos better is noise in highlights, right? Highlights as in what region of the histogram (if you were looking at live view) or what % on a waveform monitor? This noise might not be an issue in a dark night scene.

No.

The noise is ALWAYS going to be in the areas with the least light. That's' why we spend so much time putting light in the shadows. Both on film and in HD. Get that wrong and everything is a muddy mess.

Ah. So if one were planning to work at daringly low light levels for certain scenes, you would recommend lighting "up" then color correcting darker as opposed to just going as dark as you want in cam?

Yep, every time.

Anyone know where this "noise floor" is? I've been attempting to do some informal tests, looking at dark areas in my apartment while looking at the histogram, hooked up to a plasma TV to see the noise. Since it's hard to tell exactly what part of the picture any particular zone of the histogram corresponds to (precisely, anyway), I'm loosely concluding that to be safe, everything should be above 20% (first line of the histogram from the left edge) to be noise free. Sound reasonable?

Sounds about right to me. Not going to look so hot coming off the camera, but it's going to look GREAT in post! And welcome to the reason why when shooting narrative, most experienced people don't worry about "making it look great in the camera". Because unless you are shooting outdoors, or are using a ton of light, it's not going to be clean.

Well, we'll see. The crew may just be me, and I only have so many lights and so much time. I was really hoping to light it as close to the look as possible, and use color correction to polish it, rather than light miles away from the look and use cc to bring it there. It may end up noisy after all.

Well, light it how it's supposed to look, just 3 time as bright! When you bring your levels down in post, it will look just the same. I've attached an example of this from my last film. See the photos for how I shot versus the directors intent.

That is definitely different. Though, to be fair, judging your "bright" still on my unreliable imac monitor, that's still pretty damn dark. Especially in the shadows (near his pants area). I can see noise if I look real hard. I thought you were saying not to really have ANY deep blacks in there while shooting, and to introduce those through color correction. Pants area of the guy and the shadows beneath look pretty damn dark on here.

It was darker than I wanted it to be. But I didn't have full control. Here were my parameters:

1. Warehouse site with no windows, and a translucent ceiling
2. A single 15A plug to accomodate ALL lighting needs
3. A 20ft x 20ft space for the actors to play in
4. We would see about 280 degrees of the set so I couldn't use stands on the set.

I had to light a 20x20 with three practicals.

Welcome to indie cinema!

Let me ask you this, then:

In your original, pre-grade lighting scheme (which is how I planned to light our movie, level-wise, with the intent of probably crushing those blacks later), did you have noise issues? In those dark areas? My original idea with this movie was to light pretty dark for the dark areas, but leave a little information (maybe even more than you did), but getting everything to read, at a minimum, 20% on the histogram would actually be going a lot brighter.

Given the choice, I would have pulled up ALL the blacks you see in this. Yes, there were noise issues, though not bad. I made some tradeoffs between letting it go black or raising ISO. Also, note that this was shot at F4.0 because we only had a lens that would do F3.5 at best. The only thing we had that was faster was a 50mm and it wasn't wide enough.

This is why I STRESS to new filmmakers, BUY GOOD GLASS. I could have saved us from a ton of noise issues in that warehouse if we had some F1.8 or F2.0 glass. But given some of the shots we had to do, I was working at F4 most of the time with only three practicals.

well yes but... F4 is not a bad place to be because you got some DoF to work with rather then against. DoF at 2.0 or less can be pretty thin, and any minor errors that F4 would cover will be obvious at F2 :(. pick your poison.

yes I know, a F2 lens @ 4 will be sharper then a 3.5 or 4 lens wide open.... but since so many folks turn their detail level down, it probably doesn't make much difference. flare level may be another story though.

The poison in this case is noisy footage which will be there even if the focus and acting is perfect. Versus perhaps doing numerous takes to get the focusing right. I'll take my chances on the focus puller versus the noise reduction software! :)

This is a widely held myth. A GOOD F4 piece of glass should be just as sharp. The problem is a lot of F4 lenses are crap. So yea, the faster and more high quality lens will look better stopped down.

I'm afraid the multiples appear to be correct and someone has done the tests to prove it:

http://kriskoster.com/images/5DM2-ISO-Noise.jpg

I've done my own tests as I shoot a lot of night video and the chart appears spot on to me.
I find it fascinating that ISO 640 has the lowest noise and far less noise than ISO 100. My daylight setting is permanently set to 160.

Interesting. We may be able to borrow a camera, but the lenses I have available through girlfriend are a 1.8 50mm prime, an f4 24-105 zoom, and a (presumably f4, haven't even messed with this one yet) 75-300 zoom.

I was planning on using the prime for a few outdoor night shots, and the 24-105 everywhere else (night interiors, with lighting) unless for some reason it doesn't cut it.

PS my girlfriend does not do video, she does stills, and shoots exclusively in RAW. Do these noise issues/ISO settings apply in that world or is it a video thing?

You and AdamJ must have made a mistake in your tests, because that chart is way off. While it's true that 160 has less noise than ISO 100 in the shadows, the reason is not "because the sensor chip design was optimised for these settings and will produce the least noise" as you guessed in your blog, the reason is that three of Canon's design flaws happen to cancel eachother out at that setting. As I explained above, one flaw is digital scaling before writing the raw data, the second is too great of bit depth (canceling out posterization of scaling), the third is failing to use the correct white point at each ISO setting, which cancels out the 1/3 stop of dynamic range that would have been lost (so that 160 uses the white point that 200 should have been using).

It might be fascinating if it were true, but it isn't. 640 does not have the lowest noise.

Some things apply, but most don't. For raw stills, it's this:

• ISO 100, 200, 400, 800, 1600 are fine.
• ISO 160, 320, 640, 1250 are fine too (but not better).
• ISO 125, 250, 500, 1000 are slightly worse (1/3 stop less headroom, i.e. dynamic range).
• Underexpose ISO 1600 rather than increase ISO above that (same noise but more headroom).

its not a myth. in fact I'm getting ready a lens review which will clearly show it. there must be another 30+ years worth of lens reviews in Pop or Modern Photography with charts showing the same. very few ( read very expensive ) lenses have very good performance wide open. close them down even one stop, never mind two and you are at the sweet spot of the lens's performance.

Now all of that said, shooting video on a dslr requires a lot less of a lens then does shooting stills. Some fairly poor still lenses will work just fine for video purposes because all the data being so down sampled.

MY personal quest right now is to find a lens I want on the camera all the time, to cover 60% of my shooting needs. I just got a Vivitar 35-105 3.5, ( there is a lot to like about this lens ) but now this lens has me seriously considering an olynpus OM 35-105 2.8. The OM is just the right specs, but the only one I've seen is with a dealer in hong kong for $1200. I'd certainly want to try it out first before....

of course new modern zooms generally do better then older ones. The kit 18-55 canon every one picks on, while slow, it still a very decent performer optically. thank you modern coatings and glass formulations.

It is -- I agree with Perrone Ford.

Here are a few examples where fast lenses stopped down are the same or worse compared to slow lenses wide open (or not stopped down as much), sometimes even when comparing huge price differences.

50mm f/1.2 ($1,500) @ f/2.8 vs 50mm f/1.8 ($100) @ f/2.8

16-35mm f/2.8 ($1500) @ f/5.6 vs 17-40 f/4 ($700) @ f/5.6

28mm f/1.8 @ f/2.8 vs 28mm f/2.8 wide open

100mm f/2 @ f/2.8 vs 100mm f/2.8 wide open

That does not necessarily conflict with what Perrone is saying. For the same cost as a fast lens that gets sharp when you stop down 1-2 stops, you can make one that is 1-2 stops slower that has even higher quality wide open.

Daniel, I wasn't going to bother. But thanks.

Huh ????? I'd looked at all those examples, welcoming to be proven wrong, but that simply wasn't the case. every single one of those lenses was softer wide open then closed down 2 stops. the 100 2.8 came the closest to being **almost** as sharp wide open as closed down, but there still was a difference, wider being softer. its certainly a very good lens overall. the 50 1.2 was particularly soft wide open, and didn't look very good until 2.0.

as for your last comment about making a sharper but slower lens, sorry by physics is in the way. when you close the iris down about 2 stops, it blocks stray light from passing thru the lens ( basic explanation ) resulting in increased contrast & resolution. sure there are some really expensive examples that might break this basic rule, but they are few and far between, and none of the lenses shown here. with very careful lens & coating formulation, you might get close, but physics is going to win in the end. the 100 2.8L was probably about the closest to having no major practical shooting difference wide open to stopped down a bit, but thats one lens. A long lens that wouldn't be your day to day shooting glass for most video purposes.

Most folks want a zoom for their goto lens that stays on the camera most of the time, which will cover a decent amount of their shooting needs. If you are really concerned with the sharpest lens, then use all primes. primes will perform better then zooms because you are again going head to head with physics that you can't cheat on. I'm not saying that modern zooms ( basically anything in the AF era ) isn't a much better zoom then the older glass, it almost always is. However, modern primes have also benefited from better glass & coating formulation. Where you might find a zoom out performing a prime is if you take a modern zoom against an older prime - a MF one.

if you want a lens sharper, you have to make it bigger. the **general** rule is that two lenses of otherwise the same specs ( focal length / stop ), the one with the larger front element will perform better. this is again all based on physics. can apo / FL glass help ? sure, but again I'd say compare 2 APO glass lenses side by side. of course the bigger the glass elements, the more they cost, so lens makers pick their compromises carefully.

most glass performs best starting at 2 stops closed down, thru about F8. Once you get smaller then F8 you start to get a pinhole lens effect from the iris again starting to soften the image. some glass may hold out to F11, but by F16 it starts to become apparent.

with all of that said, sharper isn't always best for video purposes with these cameras. the slightly ( sometimes not so slightly ) softer older glass still makes a nice video image, because the softer image produces less moire / aliasing. so there is clearly a trade off point here. in practical shooting terms though, wide open may produce visible softness that 2 stops in won't, regardless of lens, which is visible when shooting video. with all things being equal in that respect, a faster lens ( 2.8 @ 4 ) will consistently be sharper then a F4 lens wide open due to the basic physics involved with how lenses work, assuming reasonable equal glass - you aren't comparing a $5k or more lens to a consumer / pro photography lens for $500 or even less.oh.... and just for the record some one has done lens tests between cine style glass and still photography glass and basically the quick answer is, the cine glass wasn't any sharper then the still lenses. However, the cine lenses have much better mechicals for shooting purposes.

We may have a failure to communicate. :) It probably didn't help that I posted the examples without any explanation. Let me try to rectify that now by explaing what I meant the examples to show, and you can tell me if it helped or not:

When you open the first link, it shows three images from the $1,500 lens (Canon EF 50mm f/1.2 L) stopped down to f/2.8:

http://thebrownings.name/misc/2010/2...8-22-06-34.jpg

If you mouse-over the images, it replaces them from three images taken from a different lens, the Canon EF 50mm f/1.8, which only costs $100:

http://thebrownings.name/misc/2010/2...3-09-27-38.jpg

Do you agree that this image is sharper? To me it is. And it is from a lens that costs over an order of magnitude less, and it's only stopped down by 1.3 stops -- much closer to wide open than the $1500 lens which is stopped down 2.3 stops.

Now let's look at the second example:

Here is the $1,500 Canon 16-35 f/2.8 II at 16mm f/5.6 (stopped down 2 stops) - mid-frame:

http://thebrownings.name/misc/2010/2...9_07-55-42.jpg

Compare that with the $700 17-40 f/4 at 17mm f/5.6 (stopped down just 1 stop) - mid-frame:

http://thebrownings.name/misc/2010/2...3_08-37-55.jpg

Do you see how the slower lens is sharper? This despite the fact that it's half the price and is one stop closer to being wide open.

I hope that helps to explain why I think the examples help support the position I've taken. In any case, I can at least say that we agree on this much:

• Usually, 1-2 stops down is sharper than wide open.

I think this is what we disagree about:

• A fast lens stopped down by two stops is sharper than a slow lens stopped down only 1 stop.

Yes, and I agree, because I agree that stopping down makes lenses sharper. But my point is something different -- that a slow lens can be (and often is) sharper than a fast lens stopped down to the same f-stop.

One of my pet peeves is when people throw the word "physics" around incorrectly. If you look at angular resolution, sure, fast f-numbers have more angular resolution than slow ones -- but that that's only in the object plane, which is useful to astrophotographers and the like with their wide converters ("field flatteners"), but not the image plane, which is how most folks think.

For a given cost, you can polish a slow lens to a higher Strehl ratio than a fast lens, if not for any other reason then at least because the faster lens is going to have far more area, and lens polishing cost scales exponentially with area.

Well, to my understanding that would only explain why it has less flare when stopping down -- not better sharpness or resolution (I think those are improved for entirely different reasons).

I think that you get less flare from not having the glass there in the first place than by putting an aperture stop to block the glass (though not all of it's reflections).

Would it be alright if we left physics out of it? At least until you have a specific physical law that you want to use to support your position?

No. Again, that only applies to angular resolution, i.e. resolution in object space. That is how astronomers and astrophotographers such as myself think about it. But photographers and videographers aren't interested in arbitrary angles of view -- they want specific wide, normal, and telephoto fields to go with their accustomed perspectives. And even if they did know about angular resolution, they don't have unlimited flexibility in backfocus distance (39mm -- it's the law!), so they couldn't convert object space resolution into image space with a wide converter anyway, because then it wouldn't focus on the camera.

That is not a general rule. Take one of the sharpest photographic lenses in existence: Coastal Optics 60mm f/4 Macro. You can take any 50mm or 60mm lens you want from f/1.0 on up, stop it down to f/4, and it's still not going to bear the performance of that lens, despite the fact that the Coastal Optics front element is much smaller. With the $4,000 cost of the lens, the lens designer, Brian Caldwell (brilliant fellow by the way), had all the freedom he wanted to make it an f/1.4, f/1.2, or f/1.0, but he didn't because he put that cost into the sharpness instead.

Agreed. (BTW, the proper term is diffraction.)

Not so. Diffraction affects all lenses equally - no lens can "hold out" any better than another.


Agreed.

Interesting.

It's not entirely true that I guessed it. I was trying to listen to the Canon rep who was explaining the reasons to us, likely I misunderstood that explanation. I'm not a sensor expert. Although Canon already acknowledge that lower noise levels can be found at higher ISO numbers. I just couldn't find another source anywhere else online (other than that chart) that could break it down properly.

My own tests found that the graph appeared to be telling the truth, although I admit my tests were mere casual observations of footage taken at various ISO levels.

I am more than happy to change the information in the blog to factual analysis. If you could point me in the right direction as to where you acquire that information from, I'd be very grateful. As a filmmaker, I would be particularly keen to learn from you what the most ideal ISO vs Noise levels are.

Kris

ARe we saying the whole thing is nonsense? Or just that the chart is inaccurate?