Scarlet X DOF, a case for small sensors.

[Zach Swena]

Hi,

I have observed that many here are sad about the absence of a 2/3" scarlet. A S35 scarlet has always been in the works in various forms since the 2008 announcement of the entire epic/scarlet line. Scarlet X is the realization of this and is a necessary part of the product line. As I see it, it seems that the low yield on ASIC production is to blame for removal of the 2/3" model. The smaller sensor was chosen, if I remember right, to reduce production cost and enable the lower price point. The original 3k 120fps camera would have required lots faster ASIC's then Scarlet X, and the price for the version Ted was running around with would have been as much or more then Scarlet X less the difference in sensor cost. I hope that when they get the chip production process refined further they will be able to revisit the 2/3" design. Here is an idea, make a 2/3" epic and 2/3" scarlet related in the same manner as the X siblings.

Why would a small sensor epic make sense? There are many professionals who make a living off of multi camera event shoots and one man run and gun shoots. For these types of setups a deep DOF is a must. There is no AC on set to maintain focus, and many times there are no rehearsal as it is a live event. Those that need the frame rate can pay the extra cost. It would be good to use a sensor size that would maintain deep DOF, but leave the option for 4k sensors in the future. The question of RED only making cameras for professionals has been answered. Now the question is whether RED is only a professional cinema company, or are they willing to let the professional video/broadcast crowd enjoy their technology too. For now, most in this area would be happy with a motion resolution that debayers to 2k or 1080p.

Those that don't need the higher frame rate could be paired with the lower end of the yield and get the scarlet version. This would significantly reduce the cost of multi cam setups.


In the mean time, is there a way Scarlet X can be handled to work in this setting?

For me the requirements are this:

I need DOF of 10-20ft min. Camera distance is not something that can change, I need full person, to upper body shots from 40-80ft away from the subject. The reason for the deep DOF is the subject needs to be able to move around unscripted on a platform. The camera operators currently just have to set focus when they move from one side of the stage or adjust iris for audience shots. If they have to continuously pull focus, then another camera must be used.

There are other examples where a smaller sensor would be preferable, but this is just the example I have.

I would love to be convinced that s35 could work for this, only then I would have to track down the money for the cameras.

 

[AndrewThomas]

Shoot windowed. 2k on Scarlet X is pretty close to 2/3. 3k is between 1" and 4/3. Mix that in with stopping down the lens a bit.

 

[Carter Cammack]

Does DOF increase as you window down the sensor? I know that the lens crop factor does.
Ferzample, a 50mm standard lens equals 80mm at 4K cinema motion, at a ratio of 1.6.
At 3K, that lens equals a 100mm lens at a 2.0 raio, and at 2K, a 3.24 ratio makes your 50mm equal to a 162mm from what I understand. At 2K, you're using less than 1/4 of the sensor area, the very center.

But does DOF increase in that same proportion?

A 3X to 4X increase in focal plane depth would be a HUGE factor, and possibly controllable with the iris.

 

[Yoav Yerushalmi]

Does DOF increase as you window down the sensor? I know that the lens crop factor does.

yes. Depth of field is a function of subject distance, focal length, aperture, and 'circle of confusion'. The circle of confusion is basically how 'refined' a pixel is, and is directly related to sensor size.

So the smaller the sensor, the deeper your focus becomes (yes this is somewhat overly simplified, but it's a valid generalization -- a lot of people preferred to shoot on the 5D over the 7D or vice versa precisely because the size of the sensor affects depth of field)

 

[Pawel Achtel]

Do you know that you can step down and achieve the same DOF on S35 frame without losing light gathering capacity as compared with a smaller sensor size? In other words, a bigger sensor gives you more options, including stepping down for higher DOF. Conversely, smaller sensor is limiting the amount of light it can gather and you would need to use super fast lenses, which do not exist, to achieve shallow DOF on smaller sensor.

Also, with the small pixel size, your resolution tends to be diffraction limited and you need to shoot at wider apertures to avoid this.

Larger sensor is the right move.

 

[Eki Halkka]

Do you know that you can step down and achieve the same DOF on S35 frame without losing light gathering capacity as compared with a smaller sensor size?

Yes and no. The total amount of captured light is the same, but the amount of light gathered by each photosite (exposure) is only about 25%.

When you compare i.e. 4K and windowed 2K, you need to stop 4K down about two stops to get the same DoF. In other words, you need four times higher ISO number for the same exposure. I.e. ISO 800 for 2K means ISO 3200 for 4K. Downscaling the resulting underexposed 4K to 2K reduces noise by averaging pixels, which is the reason for that partial "yes", but it will not introduce new detail to areas where the sensor sees none, areas that are more or less fully in noise. And this is the reason for the "no".

Plus, there's the issue with glass: there's no affordable, light and compact long range zooms for the S35 format.

 

[Pawel Achtel]

We have discussed this about 2 years ago, if I remember correctly. Four times higher ISO downscaled from 4k to 2k will give you the exact same noise level all other things being equal, which would make it yes and yes

However, I do agree that there are no equivalent lenses to, say a 40x or 44x zooms, like Canon and Funinon make for 2/3" 3CCD prism sensors. But fixed Scarlet could not use those anyway. A changable 2/3" Scarlet could not use those lenses either because they do not focus all colours on one plane. They are designed for 3CCD prism system. It would require an optical adapter of some sort. Also, I wouldn't call the long range zooms, like a 40x zoom eng lens affordable. If you are into sports or bird shooting, I think you may need to stay with a 3CCD ENG camera simply for the lack of lens choices.

Also, neither Scarlet nor Epic have loop recording (cache), which is a step back from Red One BTW, and which you would probably want for either sports or bird shooting.

 

[Eki Halkka]

We have discussed this about 2 years ago, if I remember correctly. Four times higher ISO downscaled from 4k to 2k will give you the exact same noise level all other things being equal, which would make it yes and yes

Only problem with the above is the fact that it does not work at all like that in real life.

And whaddya know, i happen to have an example of this exact thing to prove my point. We shot some tests with Epic, to see how much it can be over- and underexposed, among other things. We shot the exact same scene, one exposed at f8, the other at f22. Two stop difference, just as when matching the DoF.

I adjusted the shots in RedcineX to match each other brightnesswise. I then scaled down the underexposed one to 2K resolution, in photoshop, using bicubic interpolation which averages the pixels (thus reducing per-pixel noise as we know). The other shot was kept at full resolution.

Here's the result at 100% pixel size, cropped for proper display on the forum (the area with red border is f22, scaled down):

...as you can see, they are hardly "exactly the same".



Here's the original 5K exports as tiff grabs from Redcine X (the upload takes about 15 minutes):
http://eki.pp.fi/digivideo/RED/A004_C007_0810G5_001.0000013.zip

 

[Pawel Achtel]

Eki, There are 4 stops between f/5.6 and f/22. 4 stops is 2^4= 16x. So the full sensor is getting 16 times less light.
The 2k crop is getting 1/4 of the light. Of course, the f/22 image will be more noisy! Duh, you are exposing it with 1/4 of the light as compared to 2k crop. Try f/11.

 

[Carter Cammack]

I'm wondering if RED or a lens manufacturer could make a "Scarlet" scalable cinema prime set, with various adaptors for the various capture settings. (Kind of the opposite of a 1.4x teleconverter.) With smooth stops and focus gears, only fully electronic for data.
When we need to scale or window down to 3k or 2k, just change an adaptor or extension tube. Three lenses, (10mm, 20mm, 30mm) and 3 ratio adaptors (2K, 3K, 4k) would give us a lot of options.

Just don't try to cover 5K with them!

 

[Eki Halkka]

Eki, There are 4 stops between f/5.6 and f/22. 4 stops is 2^4= 16x. So the full sensor is getting 16 times less light.
The 2k crop is getting 1/4 of the light. Of course, the f/22 image will be more noisy! Duh, you are exposing it with 1/4 of the light as compared to 2k crop. Try f/11.

My bad... you're of course right, it wasn't a two stop difference. I blame late night for my brain fart ,-)

But it wasn't a four stop difference either. The shots are f8 and f22, not f5.6 and f22. Also, scaling was done from 5K to 2K, not from 4K, which biases the evaluation a little (by 20% i guess), favoring the more underexposed f22 shot. We didn't do all stop variations, when i have time i'll try to see whether we did f11, i recall not.

 

[Eki Halkka]

i'll try to see whether we did f11, i recall not.

Nope, we did not. So here's the next best thing: the f22 shot scaled to 25% (about 1K) size. This should in theory, following the same logic, compensate for a four stop difference in noise performance and the scaled down version should look "a stop better" noise wise than the f8 shot. It does not... so my original point still stands.

Next time i shoot Epic, i'll do the actual test with a two stop difference, if i have the time. That might take a while though, i rent cameras on a case per case basis with my customer's money, and more often than not the choice is something less than Epic...

PS: I have a hunch for the reason the reality doesn't follow the theory: it's due to the bayer pattern. The color noise of a debayered image is larger than one pixel in size because color information isn't captured at full resolution, and scaling it down doesn't really merge the color noise together, only makes it smaller. It's only after the image is scaled down more than to 50% size - where also the red and blue channel have an unique sample from the sensor for every pixel - that averaging pixels really starts to bite the chroma noise as predicted.