In camera stitiching

[wofsy]

I wonder how good in-camera stitching is in digital cameras. Are any any good?

I saw the article on the Seitz 6x17 but this is out of my affordability range.

[LesPalenik]

Seitz is in a class by itself - very fast and precise.
But if you use Gigapan Epic Pro or a similar rig in combination with Autopano or Hugin, you can get comparable results.

[Telecaster]

The best in-camera stitcher I've used runs on my iPhone.   

-Dave-

[Jim Kasson]

Seitz is in a class by itself - very fast and precise.

I'd never heard of this camera. I'm currently doing slit scans:

http://www.kasson.com/galleries/timescapes.php

with a Betterlight Super 6K. I put the camera on a tripod, and lie to it and tell it it's on a rotating platform. The fastest it will scan is one pixel column in 1/300 sec. It looks like the Seitz is about ten times as fast -- 1/2000 sec per pixel (I'm assuming they mean pixel column). But ten time as fast doesn't square with doing a whole full-res  scan in less than a second, unless a full res scan is 7500 pixels high by 2000 pixels wide. What am I missing here?

I've been exploring conversions from 4K video captures, but this could be a possibility. The video captures don't get to 2000 frames/sec, but you can use 10 to 100 pixel-wide swaths.

I see the Seitz only works tethered, like the Betterlight. Ugh.

Do you see a way to fool the Seitz like I'm doing with the Betterlight?

Thanks,

Jim

[BernardLanguillier]

The samples I've seen from the Seitz were displaying really poor pixel quality IMHO.

Far worse than those of recent DSLR costing 10 times less.

The fundamental issue with scanning technologies, at least on the Betterlight Super6K that I own, is that the longest available scan line time is too short to be able to gather enough light when conditions get a little bit dark. This forces to raise the ISO and images become quickly way noisier than those of recent DSLRs at base ISO. Yes, you get true RGB pixel, but the problem you create is IMHO much worse.

IMHO, following many first hand testing with expensive equipment I paid for myself (and that I really hoped would work better), nothing beats spherical stitching for real world applications.

Cheers,
Bernard

[Jim Kasson]

The samples I've seen from the Seitz were displaying really poor pixel quality IMHO.

Far worse than those of recent DSLR costing 10 times less.

The fundamental issue with scanning technologies, at least on the Betterlight Super6K that I own, is that the longest available scan line time is too short to be able to gather enough light when conditions get a little bit dark. This forces to raise the ISO and images become quickly way noisier than those of recent DSLRs at base ISO. Yes, you get true RGB pixel, but the problem you create is IMHO much worse.

That's depressing. If you wanted to do high-speed slit scans, would you consider processing video files to be the most promising option? I'm currently waiting on an external recorder for the Sony a7S to do a proof-of concept test. The industrial slit scan camera that I know about doesn't really have the resolution I'm looking for.

Jim

[Jim Kasson]

The fundamental issue with scanning technologies, at least on the Betterlight Super6K that I own, is that the longest available scan line time is too short to be able to gather enough light when conditions get a little bit dark. This forces to raise the ISO and images become quickly way noisier than those of recent DSLRs at base ISO. Yes, you get true RGB pixel, but the problem you create is IMHO much worse.

What I do about that with the slit scans, is just average more, oh, let's call them "horizontal" pixels to get one output pixel. I have compressed 1,000,000x6000 (actually, 16 64000x6000 files) pixel files to 6000x6000 that way. The noise just disappears.

Too bad you can't so that when the diode array moves.

Jim

[LesPalenik]

The samples I've seen from the Seitz were displaying really poor pixel quality IMHO.

Far worse than those of recent DSLR costing 10 times less.

The fundamental issue with scanning technologies, at least on the Betterlight Super6K that I own, is that the longest available scan line time is too short to be able to gather enough light when conditions get a little bit dark. This forces to raise the ISO and images become quickly way noisier than those of recent DSLRs at base ISO. Yes, you get true RGB pixel, but the problem you create is IMHO much worse.

IMHO, following many first hand testing with expensive equipment I paid for myself (and that I really hoped would work better), nothing beats spherical stitching for real world applications.

Cheers,
Bernard

Seitz has been making rotational cameras for many years, both film and digital.
Their Roundshot Super film cameras processed the scene in two passes. During the first pass the camera measured the available light along its path, and in the second pass as the camera turned, it adjusted its speed according to available light. The light compensation worked quite well and since it was completely analog, most of the time there was no banding.
 
With their Roundshot Super 220 or 35, using a long lens, it was possible to create a very detailed, an almost 2m long film panoramic frame.
Using a wide-angle lens, the frame length for 180-360 degrees was measured in inches. The shorter frames similar in length to 6x17 could be easily scanned. You didn't have to do any stitching, however to make prints from those long frames one required a special Roundshot enlarger. I found the image quality and workmanship exceptional, especially when using the 220 model.

[BernardLanguillier]

Seitz has been making rotational cameras for many years, both film and digital.
Their Roundshot Super film cameras processed the scene in two passes. During the first pass the camera measured the available light along its path, and in the second pass as the camera turned, it adjusted its speed according to available light. The light compensation worked quite well and since it was completely analog, most of the time there was no banding.
 
With their Roundshot Super 220 or 35, using a long lens, it was possible to create a very detailed, an almost 2m long film panoramic frame.
Using a wide-angle lens, the frame length for 180-360 degrees was measured in inches. The shorter frames similar in length to 6x17 could be easily scanned. You didn't have to do any stitching, however to make prints from those long frames one required a special Roundshot enlarger. I found the image quality and workmanship exceptional, especially when using the 220 model.

All that is of course very true, but it reminds me of another "problem" with these cameras, they can only do cylindrical projections.

Cheers,
Bernard

[LesPalenik]

Very true, Bernard. Their older cameras especially the 35mm models armed with medium or long lenses could produce only the classic one-row "skinny" panoramas. However, using the 6cm (120/220) film and a 20-35mm lens, the resulting film frame produces quite a substantial file that when done in modern digital method would require quite a few single exposures in two or three rows. In many situations it is no hardship and it can be easily achieved at a much lower cost with stitching multiple segments, but for aerial or some action photos, the single shot panoramic capture is still preferable.

However, Seitz recently announced a new "Metric" model that uses Canon 6D with a fixed 20mm lens on a modified VR drive (another Seitz product) that takes multiple shots in 5 rows and produces a combined file, almost 200MP in size. Very similar approach as you are employing under manual control, however, the Metric's operation is fully automated (both for picture taking and subsequent image stitching), and the camera, lens, and rotational gear are precisely calibrated for one-pixel alignment.

Using the standard Seitz VR drive, one can use practically any late camera for production of semi-automatic panos, but not with the same speed and accuracy as the new Metric model.

[LesPalenik]

I'd never heard of this camera. I'm currently doing slit scans:

http://www.kasson.com/galleries/timescapes.php

with a Betterlight Super 6K. I put the camera on a tripod, and lie to it and tell it it's on a rotating platform. The fastest it will scan is one pixel column in 1/300 sec. It looks like the Seitz is about ten times as fast -- 1/2000 sec per pixel (I'm assuming they mean pixel column). But ten time as fast doesn't square with doing a whole full-res  scan in less than a second, unless a full res scan is 7500 pixels high by 2000 pixels wide. What am I missing here?

I've been exploring conversions from 4K video captures, but this could be a possibility. The video captures don't get to 2000 frames/sec, but you can use 10 to 100 pixel-wide swaths.

I see the Seitz only works tethered, like the Betterlight. Ugh.

Do you see a way to fool the Seitz like I'm doing with the Betterlight?

Thanks,

Jim

Jim, I don't know if Seitz 617 camera is scanning only 1 pixel column or multiple pixel columns at a time. I would guess only 1 pixel wide, but possibly with some buffering. Their older film models operated all on a narrow slit concept using the best (center) part of the lens.

In addition to the 617 model that is now several years old, they have also a new rotational all-weather Livecam model with 2048 vertical pixels and a similarly specced out portable D2X model. D2X works in a standalone untethered mode, whereas the Livecam model can feed directly into a server that creates continuously new pano images. Both of these cameras use fixed or inexpensive zoom Nikon DX or FX lenses, and offer 9 stops of DR, and 100-800 ISO range. They cover up to 360 degrees, with one degree increments, so one degree width will encompass several pixels.

[BernardLanguillier]

However, Seitz recently announced a new "Metric" model that uses Canon 6D with a fixed 20mm lens on a modified VR drive (another Seitz product) that takes multiple shots in 5 rows and produces a combined file, almost 200MP in size. Very similar approach as you are employing under manual control, however, the Metric's operation is fully automated (both for picture taking and subsequent image stitching), and the camera, lens, and rotational gear are precisely calibrated for one-pixel alignment.

Interesting.

I have been toying for some time with the idea of buying this: http://www.dr-clauss.de/en/foto-studiotechnik-3/rodeon-pix-series-eng-3

It is expensive, but is a totally generic solution, able to do robotic panorma with lenses as heavy as my 400mm f2.8 on the D810, controllable by an iphone using a local wifi network generated by the head.

Cheers,
Bernard

[LesPalenik]

Thank you for posting that link, Bernard! Interesting in many ways.