10 Pixel Offset

[Phil Bates]

I read on an old thread that there can be a 10 pixel offset difference(?) between any two Red Cameras, which can make S3D capture challenging. Unfortunately, there was not enough information for me to understand what this is about.

Can somebody share with me what this is? Is it an alignment of the sensor?
Is it really as big a problem as some say it is?
Can this offset problem be fixed in post by manually aligning right or left side?
Any other information you can add would be helpful.

Thanks,
Phil

 

[Jonathan Smiles]

If your shooting with primes in 3D it's not a big issue as when you align the rig you take into account the sensors and lenses.

However if shooting with zoom lenses in 3D then having both zoom lenses track accurately is very important as you can't re-align the rig for every focal length.

The answer is to shoot with zoom lenses that can be adjusted for tracking, the most common in 3D being the Angenieux short zoom. This procedure is not trivial and will result in the lens being paired with the camera it was adjusted for.

 

[Phil Bates]

If your shooting with primes in 3D it's not a big issue as when you align the rig you take into account the sensors and lenses.

However if shooting with zoom lenses in 3D then having both zoom lenses track accurately is very important as you can't re-align the rig for every focal length.

The answer is to shoot with zoom lenses that can be adjusted for tracking, the most common in 3D being the Angenieux short zoom. This procedure is not trivial and will result in the lens being paired with the camera it was adjusted for.

Thanks Jonathan. I forgot to ask, is there a way to measure the 10 pixel offset to see if it really exists between two cameras?

Thanks again,
Phil

 

[Jeff Kilgroe]

Some collimation tools can give you measurements. The Optimator from Optitek has measurements in its reticle image so you can adjust your FFD [backfocus] and also see how far off-center the sensor is. I don't recall how precise those measurements are. Other tools like the NULL target and those with a normal siemens star will show you the camera is off, but not give any precise indicator of how much. You can get a more pixel-precise number by shooting with two cameras in a beam splitter rig with a zero-convergence as you would use for shooting HDR or other multi-exposure processes. Also keep in mind that differences between cameras/sensors, etc.. can be both horizontal and vertical.

Once again, not a RED problem, but a general camera thing. There are certain limitations to tolerances in any manufacturing process and it's not uncommon for sensors (or film gates) to be off just a bit between cameras. At least sensors don't move after the fact though. Unlike shooting 3D with film cameras where you can have other factors such as gate weave and flexion of the film itself to deal with.

You align your rig for each focal length. It can become an issue when using zooms as Jonathan points out. I think ET's Quasar 3D rig can be configured to compensate for tracking as part of its convergence control. But I have not shot with one to know for sure. Vertical differences can be accounted for by shimming the camera mounting, but this can be more involved and will also differ with focal length, so much harder to deal with in terms of how most rigs are constructed. Lenses can vary in their centrical alignment as well, but typically adhere to higher tolerances -- optical elements are less forgiving if slightly out of alignment compared to a film gate or sensor.

Tweaking IO and C in post can usually cover most such alignment issues as they're very small and usually unnoticeable except when zooming. In extreme cases, if the sensor seems way out of alignment, I would see if RED can fix it. If you zoom and it's obvious that the center of your sensor doesn't remain lined up while zooming, just by watching a rack zoom, then you definitely need to get that checked.

Upcoming DSMC cameras from RED will have user-adjustable back-focus and another multi-axis adjustment via factory or service center. So there's a much better chance of having all your Epic cameras dialed to be nuts-on.

 

[jonnycom]

My understanding is if the offset is horizontal it can easily be fixed in post by changing convergence, but if it's a vertical offset it's more critical that you fix it with the rig.

it's also my understanding you cannot change IO in post, although a lot of post tools claim that they can do that.

 

[Phil Bates]

Thanks guys!

 

[Thor Melsted]

it's also my understanding you cannot change IO in post, although a lot of post tools claim that they can do that.

You can up to a point. It's limited in how much of an adjustment you can do, depends on a lot of factors, but primarily the complexity of the subject matter.
The more overlap there is, the worse it will work. On green screen with simple subjects or shots of wide open spaces it can work extremely well.

I'm just guessing here, but I think the processing principle must be similar to speed changes, like with Twixtor - where it simply considers the two frames as part of a sequence and interpolates "motion" between them to create extra frames, and then gives you those frames as new IO options.

 

[jonnycom]

Hi Thor,

Can you change the position of a single camera in post on a single camera shoot? Because that seems like the same thing with changing the IO.

I know you can do things like projection mapping and other tricks to simulate camera movement in post but in stereo it seems like it would be impossible as you can't really change the camera perspective.

I have a few programs that say they are changing IO but I don't think so.

 

[Jeff Kilgroe]

Changing IO in post is possible to an extent, but it can be very involved and becomes exponentially more complex as the amount of correction increases. You can only warp perspective and crop into frames so much before it becomes a full on 3D/CGFX job to rebuild a shot. There's software out there now that claims to be good enough where we can shoot side by side with large cameras like the RED One and correct IO distance in post. Hmmm.... I've seen a bit of this footage and I'm not really convinced. It seems to work at longer focal lengths and subject distances where convergence can be used to compensate, but I really haven't seen enough of that footage from this newer software to know for sure.

Generally speaking, changing IO in post (or attempting to do so) is not very practical, or even probable beyond very minor adjustments.

 

[Jeff Kilgroe]

Changing IO in post is possible to an extent, but it can be very involved and becomes exponentially more complex as the amount of correction increases. You can only warp perspective and crop into frames so much before it becomes a full on 3D/CGFX job to rebuild a shot. There's software out there now that claims to be good enough where we can shoot side by side with large cameras like the RED One and correct IO distance in post. Hmmm.... I've seen a bit of this footage and I'm not really convinced. It seems to work at longer focal lengths and subject distances where convergence can be used to compensate, but I really haven't seen enough of that footage from this newer software to know for sure.

Generally speaking, changing IO in post (or attempting to do so) is not very practical, or even probable beyond very minor adjustments.

 

[Thor Melsted]

Hi Thor,

Can you change the position of a single camera in post on a single camera shoot? Because that seems like the same thing with changing the IO.

I know you can do things like projection mapping and other tricks to simulate camera movement in post but in stereo it seems like it would be impossible as you can't really change the camera perspective.

I have a few programs that say they are changing IO but I don't think so.

Jeff is absolutely correct.

Changing the angle or position of a single camera shot means you have a narrow window of minor mods before it becomes a full on visual effect - as in the case of 3D conversions.
But in a dual camera setup you actually have two angles with which you can interpolate a third - which allows you to change the IO somewhat, although it is still pretty limited. But it's a far cry from creating a whole new angle from a single source.

Under perfect conditions IO adjustment works - but conditions are rarely perfect unless you work hard at making them perfect - in which case you'd probably be using a beam splitter anyway.

Here's a demonstration of The Foundry's Ocula - using a very complex shot (shot on two parallel RED Ones). They show off IO modification starting at around 3:45.

I've also seen the RE:flex approach, and I've tried it out. It works OK with the correct footage, but again, I wouldn't consider it a permanent or even practical solution.

So to answer the question, yes, it's possible, but it depends on how much time and money you're willing to invest in it, and whether that money isn't better spent getting a rig more suitable to your shoot.
That being said, supposedly Ocula was used to fix IO in multiple shots in Avatar, so I guess there's a need for it even if you use the best rigs.