Pixel shift with stereoscopic imagery

[fde101]

The whole concept of a "crop factor" is kind of a bad idea, because it leaves out the question of what you are cropping FROM.

You don't really "crop" the lens, nor do you ever capture the entire image projected by a lens that covers the format you are shooting in.


What Peter is getting at is that the sensor in the Scarlet, when capturing 4K, is the same size as a Super35mm film frame -- the size of a typical 35mm film MOVIE camera, which is smaller than a FF35, 35mm STILL camera.

So there is no "crop" from the field of view of a MOVIE camera at 35mm, but there IS a "crop" from a FF35mm STILL camera -- much like a 35mm film still camera would have a huge crop factor if you were comparing to a 4x5" view camera, as would even a Hasselblad or other medium-format camera.

Is a FF35 a "reverse crop" from S35, or from micro-4/3? No, it's just a different format. The "crop factor" is mainly a tool to help people who are used to one format convert the lens focal lengths they are used to into some other format when they start shooting on it.

ALL cameras "crop" the image from the lens -- at least I haven't seen one that captures the circular image they project. It's more a question of how much of the image they actually capture.

A 50mm lens will give the same image coverage on a S35 camera whether that lens was designed to cover S35, FF35, or medium or large format. Stick the same lens on a micro-4/3 camera, and it will capture a smaller part of the same image. Stick the lens on a 2/3" camera, and it will capture an even smaller part of the same image.

They are ALL cropping the image, however.


EDIT: This brings up an interesting idea... What about using a lens that is designed to cover a sensor larger than the one you are using, such that you get the sharpness from the center of the image and don't capture the less-sharp areas around the edges?

[JonFairhurst]

Yeah, it's a crop factor in the 35mm photo world, but it's the normal size in the 35mm film world - where the film moves vertically, rather than horizontally, through the gate.

I would think that the S35 (1.6x crop of full frame) would be an advantage for the project. First, you don't want shallow DOF. You want the whole frame to be in focus. Second, with S35, you would be using the sweet spot of 35mm photo lenses, avoiding soft corners.

Sorry to continue beating the drum, but rolling shutter could cause problems as you do a "high speed dolly in". I believe that most rolling shutter plugs deal with panning problems where vertical lines become diagonals. They may also deal with tilt effects - things moving up become shorter and things moving down become longer.

A dolly in is usually slow, so I doubt that the plugins deal with it. With rolling shutter, the frame is captured at different times. With a dolly in, occlusions change over time. That will cause errors in the 3D field of view.

Even worse - items that are close move radially from the center quickly while distant items remain relatively static. That means close items near the top of the screen will be squished, at the bottom, they'll be stretched, and at the sides, they will fan out toward the bottom.

Given that the project seeks to make a 3D representation, I think this needs to be corrected during the modeling process. It certainly doesn't make things simpler. In essence, not only are the frames taken at different times - the lines are captured at different times too.

[Peter Majtan]

Originally Posted by JWhitehead

No crop factor? I'm getting my data from this page:

http://nofilmschool.com/2011/11/red-...ors-datarates/

30fps @ 4k QuadHD = 3840 x 2160

1-12 fps 5K FF
1-30 fps 4K HD
1-48 fps 3K HD
1-60 fps 1080p HD
1-120 fps 1K

-Jim

The sensor is cropped - don't get me wrong - but the sensor is FF35 to begin with, which is why many of the cine lenses having hard time covering it completely without vignetting. Once you crop it down to 4K it will become standard cine S35 size - which means there is not crop-factor to affect the FOV. So on the other hand - in FF35 5K mode the cine glass is actually wider - if it can cover the sensor that is...

[Peter Majtan]

Jon - you really need to check out the RollingShutter from The Foundry - it is using pixel motion analysis - regardless of which way the motion goes (in this case from bottom of the frame toward the top - or vice versa depending on which way the camera is mounted). It is computationally heavy - but the results are amazing...

[Peter Majtan]

Originally Posted by JonFairhurst

Even worse - items that are close move radially from the center quickly while distant items remain relatively static. That means close items near the top of the screen will be squished, at the bottom, they'll be stretched, and at the sides, they will fan out toward the bottom.

I think the camera is pointed straight down, but what you have said still applies, just in terms of objects in the center vs the top and bottom of the image. Ironically (and Jim please correct me if I am wrong) - this is, I believe, what the imaging SW is using to calculate the 3D imagery...

[JWhitehead]

Originally Posted by Peter Majtan

I think the camera is pointed straight down, but what you have said still applies, just in terms of objects in the center vs the top and bottom of the image. Ironically (and Jim please correct me if I am wrong) - this is, I believe, what the imaging SW is using to calculate the 3D imagery...

The way I understand the imaging software's processes (I'm the intern assigned to finding the camera information for the grant proposal. I have some training on it, but haven't had access to our machine that is powerful enough to compute the DEM so I can get a firm grasp on the intricacies of the program) is that it uses a known set of points (We'll be running the images through GPS Phototag and using a high-accuracy Trimble GPS unit for our GPS tracklog) and the software looks at changes in perspective as the camera moves along its assigned path to determine elevation changes on a mm scale.

With regard to how the subject is moving along the sensor: It is moving from top to bottom, like a cinema camera. The image is not moving side-to-side along the sensor like a 35mm still camera's film would.

We're working on an in-house software package that will be more flexible than the commercially-available software package we have been using, so post-processing with regard to a rolling shutter may be possible (our commercial package wants .jpg images with no post-processing).

-Jim

[JWhitehead]

Originally Posted by JWhitehead

As far as lenses, we're looking at the RPP 18mm or 25mm. I read on redusers.net that the MYSTERIUM-X has a 1.3x focal length multiplier, but I haven't been able to get RED tech support to confirm nor deny this statement. Do either of you guys know whether this is a true statement?

Hello again everyone.

I still haven't been able to get an answer on this. Does anyone know if there is a focal length multiplier with RPP lenses? I'm asking because currently we're using a D5000 with a Nikkor 18-55 lens, and because of the sensor size, setting the focal length on the lens to 18mm gives us an effective focal length of 27mm (because of a 1.5x focal length multiplier).

Thanks

-Jim

[Peter Majtan]

Originally Posted by JWhitehead

Hello again everyone.

I still haven't been able to get an answer on this. Does anyone know if there is a focal length multiplier with RPP lenses? I'm asking because currently we're using a D5000 with a Nikkor 18-55 lens, and because of the sensor size, setting the focal length on the lens to 18mm gives us an effective focal length of 27mm (because of a 1.5x focal length multiplier).

Thanks

-Jim

Jim, I did try to explain this earlier. Crop-factor only applies if you are using lenses designed for different size sensor. For example if You are using still lenses on a APS-C DSLR (like your D5000 example) - there is a crop factor. Same if you would use still lenses (designed for Full Frame (FF)) on the Epic/Scarlet - you will get a crop factor (1.6x). But the RPP's are cine lenses and designed for cine standards - therefore there will be no crop-factor when using the 18mm RPP on the Epic/Scarlet in 4K resolution (equivalent of S35)...

Having said that - your FOV (which is your main concern) is affected by the sensor size. To simplify this - since the sensor in the Epic/Scarlet is similar size to the D500 - you will get approximately the same FOV with the 18mm as you would get on the D5000 (with 18mm). If You are happy with what you are getting with the 18mm + D5000 combo, you will be happy with the Scarlet/Epic + 18mm RPP combo...

[JWhitehead]

Originally Posted by Peter Majtan

Jim, I did try to explain this earlier. Crop-factor only applies if you are using lenses designed for different size sensor. For example if You are using still lenses on a APS-C DSLR (like your D5000 example) - there is a crop factor. Same if you would use still lenses (designed for Full Frame (FF)) on the Epic/Scarlet - you will get a crop factor (1.6x). But the RPP's are cine lenses and designed for cine standards - therefore there will be no crop-factor when using the 18mm RPP on the Epic/Scarlet in 4K resolution (equivalent of S35)...

Having said that - your FOV (which is your main concern) is affected by the sensor size. To simplify this - since the sensor in the Epic/Scarlet is similar size to the D500 - you will get approximately the same FOV with the 18mm as you would get on the D5000 (with 18mm). If You are happy with what you are getting with the 18mm + D5000 combo, you will be happy with the Scarlet/Epic + 18mm RPP combo...

Thank you for the response. This was much easier for me to understand what you were getting at. Our effective focal length with the D5000 was 27mm, and I specced a RPP 25mm, which should bring us slightly closer to the road surface. Thanks again, this is exactly what I was looking for.

-Jim