Sony A7r III with 169,6 MP Pixel Shift Multi Shooting

[Michael Erlewine]

Sony A7r III with 169,6 MP Pixel Shift Multi Shooting

Having used the pixel-shift in the Pentax K1, which worked pretty well for still photography, I would tentatively trust (an oxymoron) Sony to do this better. And since I only shoot still photos, this could give me huge images with (I hope) great resolution.

Although, the process of the pixel-shift technology is done in post through an app provided by Sony. That's fine with me. Perhaps with more room they will do a better job. Anyway, color me hopeful. Having four separate images (unlike one combined image) is probably a good thing.

[DougDolde]

From DPReview.  I'd say no good for landscapes, maybe studio and non moving scenes

New to the a7R III is a multi-shot resolution mode that, much like the system in recent Pentax DSLRs, shoots four images and moves the sensor between each shot, so that each pixel position in the final image is captured with a red, a blue and two green pixels. This cancels out the side-effects of the Bayer color filter array, meaning that full color information is captured for every pixel. This has a noise benefit both from capturing multiple shots of the same scene, which helps average out the noise, while also reducing the additional softness and noise that usually comes from the demosaicing process.


However, unlike the system Pentax uses or the earlier, 8-shot process used by Olympus, the a7R III cannot assemble the final images in-camera. Instead four Raw files must be processed using a freely downloadable image processing application for PCs that Sony will offer. The camera must also wait either 0.5, 1, or 2 seconds between shots for the sensor to settle, which is likely to exacerbate the problems of subject movement between the first and last shot.

[shadowblade]

This number refers to color information right? the output files would still be 42 MP or just slightly better in terms of resolution?

Yes.

To generate a true 169MP file, the sensor would have to shift half a pixel in each direction (instead of a full pixel), taking images equal to the full exposure time at each of the points in between the pixels and combining them to generate the increased-resolution composite. This would not only give you a file with four times the pixel count, but with an effective sensor area (for noise/signal purposes) of four times the real sensor area, while retaining the original sensor area for the purpose of angle of view and depth of field calculations.

Ideally, instead of four full-length exposures, it would shift rapidly, and multiple times, between the four positions, in order to average out motion blur during the exposure. Naturally, this would be combined with read noise subtraction, to counter the effects of using multiple sub-exposures rather than single full exposures.

All this is certainly achievable in firmware, or as a camera app, but I doubt Sony will introduce it. It would have to be done as a third-party app for the platform.

It would be great if Sony/Nikon/Canon embrace the idea of a web store for private and third-party app developers to sell apps for their cameras, as Apple and Android phone manufacturers have done. People could pull a lot more capability out of these cameras than are built into them natively (as Magic Lantern have done with Canon bodies). Certainly, the first manufacturer to embrace the app/software side of things is going to win a good chunk of market share from the others, and likely force the other two to follow suit.

[shadowblade]

From DPReview.  I'd say no good for landscapes, maybe studio and non moving scenes

New to the a7R III is a multi-shot resolution mode that, much like the system in recent Pentax DSLRs, shoots four images and moves the sensor between each shot, so that each pixel position in the final image is captured with a red, a blue and two green pixels. This cancels out the side-effects of the Bayer color filter array, meaning that full color information is captured for every pixel. This has a noise benefit both from capturing multiple shots of the same scene, which helps average out the noise, while also reducing the additional softness and noise that usually comes from the demosaicing process.


However, unlike the system Pentax uses or the earlier, 8-shot process used by Olympus, the a7R III cannot assemble the final images in-camera. Instead four Raw files must be processed using a freely downloadable image processing application for PCs that Sony will offer. The camera must also wait either 0.5, 1, or 2 seconds between shots for the sensor to settle, which is likely to exacerbate the problems of subject movement between the first and last shot.

Capturing four separate RAWs is actually a good thing. It provides the option of separating out the images and using them individually, or using individual images in moving areas where the combined image is affected by motion blur (if it's colour artifact that you're trying to eliminate, you can still get the noise/DR benefit by shifting each shot into its correct position and averaging them the normal way).

A combined RAW is much more difficult to separate out into its constituent images, particularly if you use a third-party RAW converter (e.g. C1 or DxO Optics), or if Sony doesn't build this ability into its own software.

So, in a way, this may actually turn out to be much more usable for landscapes with areas of subtle movement (e.g. ripples in mostly-still water) than the Pentax approach, owing to the ability to separate out the individual RAWs. It'll just take a bit of post-processing to get there.

[shadowblade]

Re: artifact from slight movement (particularly cross-hatched colours in areas of movement)

I wonder if this can be eliminated by taking multiple pixel-shift shots and averaging them in Photoshop. The motion blur will still be there (as in any long-exposure shot) but the artifact would be averaged out. It would probably take a lot of exposures, though.

Then again, you could just as easily take multiple ordinary shots and average them the usual way, without using pixel shift, and still get the benefit of increased DR, better colour and less noise, but still using Bayer interpolation rather than full RGB at each pixel.

[Telecaster]

Personally I’ve already got enough pixels. Better pixels, OTOH, do have appeal.

-Dave-

[Jack Hogan]

This has a noise benefit both from capturing multiple shots of the same scene, which helps average out the noise

Yes, but every camera out there could capture multiple shots in quick sequence, so this is really no advantage...

while also reducing the additional softness and noise that usually comes from the demosaicing process.

This latter one is indeed an advantage associated with the technique as long as nothing is moving.

Jack

[ErikKaffehr]

Hi,

Yes, 42 MP resolution. But it would have less colour aliasing. It would also havde less noise.

Best regards
Erik

This number refers to color information right? the output files would still be 42 MP or just slightly better in terms of resolution?

[32BT]

Personally I’ve already got enough pixels. Better pixels, OTOH, do have appeal.

-Dave-

Or perhaps just "more interesting"  pixels? You know, when people realise that what you point the camera at and how you frame it, is "more interesting" than infinite resolution.

[32BT]

But why go for the color info instead of the resolution?

Because of lens limitations and overlap in sampling. If you want to go that route, you'd be better served with pano stitching, which can provide ample resolution with mediocre lenses that can simply never be matched by a single lens and oversampling.

[32BT]

I already have to deal with focus stacking, adding pano stitching would be too much hassle, and it's also a close subject so I'm guessing pano would be even more complicated.
With a very good lens I can keep large apertures (since I'm doing focus stacking anyway) so I can reduce diffraction.

I meant in general. Yours is a very niche case, i think. Maybe you can still get a scanning back , or a monochrome camera with colorfilters. Highres and correct color. Maybe even multispectral imaging...

[BobDavid]

The eight-shift (half-pixel) multi-shot mode works well with the Pen F as well as the E-M5 II (no experience with the OMD E-MII).

This is a great feature for static product photography, copying flat art, and on occasion, outdoors. When outdoors, I always take a single shot photo of the same seen. Then, I have a fallback to cancel out artifacts.

The resolution bump with the Olys is great. It takes awhile to learn the ends and outs of using multi-shot.