Dynamic Range

[Daniel Jackson]

Sorry for this geeky thread, but I had an idea that seemed somewhat interesting, to me at least.

It seems clear that the future of digital video is going to be in maximizing dynamic range. The 5dmk2 is a great camera because it gets around 10 stops of range compared to the 6 or 7 stops most digital video had just a few years ago. The dream, that seems not to distant is when we can buy a consumer camera that tops the 14 or so stops of range that a 35mm film camera has. If you go on to imagine more sensitive chips going way past 14 stops, to 20 or 30 stops (I know, but maybe in 20 years) then there would never be a scenario where you didn't get the shot because of exposure. You could choose a just a part of the range or apply a curve and cram it all within the range of whatever you are outputting to. You could shoot against a window with no lights and decide later how much fill to dial in. You could go from candlelight inside to a sunny day outside.

Here is the especially geeky part:
Because of one of the quirks in the 5Dmk2 design, not all of the lines of resolution are used. The chip is capable of a 21 megapixel image but has to throw out lines of data in order to make a 1080 image. It seems that if one line was used for an over exposure then the next line could have a slight nd and the next line a little more, and you would have several exposures within the same frame. The several exposures captured could be used to create a high dynamic range photo just like in still photography. And the fact that it all happens at the same moment and through the same lens, the subject wouldn't need to keep still. ...so, someone at Canon get to work on this right away.

[Daniel Browning]

Originally Posted by Daniel Jackson

The 5dmk2 is a great camera because it gets around 10 stops of range compared to the 6 or 7 stops most digital video had just a few years ago.

The 5D2 is a great camera for a lot of reasons (price, size, lens selection, color depth, etc.), but dynamic range is not one of them. From my tests, you can't use more than 6-8 stops (depending on light/white balance) before you get pattern noise in the shadows. Video cameras from a few years ago have just as much dynamic range, if not more, because their shadow noise, while higher, was random instead of patterned. But it does have good "color depth", that is, very low noise over the parts of the dynamic range that is usable, which is often confused (e.g. in MFDB) for dynamic range.

Originally Posted by Daniel Jackson

If you go on to imagine more sensitive chips going way past 14 stops, to 20 or 30 stops (I know, but maybe in 20 years)

Having chips that are "more sensitive" doesn't necessarily improve dynamic range. What is really needed to do that is to increase the effective full well capacity.

[Martin Guitar]

Dynamic range is not limited to shadows. The 5Dmk2 shines in the upper area / highlights. That's why it's called dynamic range. Dakest to lightest. 6stops? You must be using a custom profile or have a bad 5D to achieve that result.

[Daniel Browning]

Originally Posted by Martin Guitar

The 5Dmk2 shines in the upper area / highlights.

Hardly. It's only 3 stops of headroom over middle gray. That's no great feat. If you enable HTP, you can bring that up to 4, but only by taking away one stop of shadows due to pattern noise.The only way to get more than 4 is to underexpose and boost exposure in post, but that's also limited by pattern noise.

Originally Posted by Martin Guitar

6stops? You must be [...]

In my experience, most people greatly overestimate the number of stops they think they are using until they actually measure it.

[Daniel Jackson]

I never checked it myself but I saw the figure of 10+ stops on the Zacuto shootout, also Robert Primes ASC did an evaluation.

[Daniel Browning]

Originally Posted by Daniel Jackson

I never checked it myself but I saw the figure of 10+ stops on the Zacuto shootout,

If you watch the shootout a little closer, you'll see that they really only showed about 7.5 stops of dynamic range on the 5D2: 3 over and 4.5 under. The 11.2 figure is from the backlit test chart, but that doesn't take pattern noise into consideration. If the 5D2 didn't have pattern noise, then 11.2 stops would be the correct number.

So then they do a realistic comparison to see how the shadow noise actually looks, where the deepest shadows are 4.5 stops under middle gray. No pattern noise, so we're OK. Then they use the same settings on a highlight test where the highlights are 5.5 stops over middle gray and the 5D2 just falls flat on its face.

If the 5D2 would have held the highlights in that shot, Zacuto would have proven it capable of 10 stops (4.5+5.5). But since it failed miserably, how much range was actually shown? This is actually easy to calculate since the 5D2 only has two levels of highlight headroom: HTP is about 4 stops and normal is exactly one less (~3 stops). (You can get more by underexposing and increasing brightness in post, but it's pretty clear they didn't do that from the test.) Since the Window looks a lot more blown out than just 1 stop, I would guess they didn't use HTP, which means about 3 stops. Add that to the 4.5 and you get 7.5.

[Justin Benn]

Interesting. I wonder how all of this is affected (if at all) by the picture profile used? In particular, if one was using Technicolor's Picture Profile for the 5D, would that enable a greater apparent range to be captured to which one would apply the appropriate LUT later? Or is this just not the case (ie. hardware capacity limited versus software algorithms?).

[Daniel Browning]

Originally Posted by Justin Benn

Interesting. I wonder how all of this is affected (if at all) by the picture profile used?

It's a tool for mapping the dynamic range of the in-camera raw footage into the 8-bit video file. If you want more than the default and HTP can provide, then it is pretty much the only way to use more than 8 stops of dynamic range on the 5D2. But then you run into pattern noise issues.

Originally Posted by Justin Benn

In particular, if one was using Technicolor's Picture Profile for the 5D, would that enable a greater apparent range to be captured to which one would apply the appropriate LUT later?

Yes.