R5 Pixel Shift rumor

Ali wrote:

Haven't seen this mentioned here yet, fresh rumor on top of the "major R5 firmware" rumor from a few days ago.

https://www.canonrumors.com/is-pixel-shift-coming-to-the-canon-eos-r5/

Don't know how possible this even is, but would be pretty amazing if it happened.

Any sensor with X/Y translation IBIS can theoretically shift the sensor for successive exposures, even when doing corrections (although shifting works best with high physical camera/lens stability).

Victor Engel wrote:

Ali wrote:

Ali wrote:

Karl_Guttag wrote:

1. whether it is true
2. whether they deal with motion in the scene,
3. whether they deal with camera motion, or if it requires the camera on a tripod,
4. whether the compositing is done within the camera or with external software (like the deal with focus stacking), or it is an option.
5. whether they save RAW or JPEG or if it is an option
6. how many frames save, or if that is an option. More frames would support dealing with more complex motion (camera or scene).

Good questions. #5 would certainly be a concern for me (I like to have RAW for all cases)

I guess thinking about this a bit more, even if Canon did add Pixel Shift to the R5, it seems highly unlikely they would add a new RAW mode for the new pixel count images that would generate.

What new pixel count? If they are shifting one pixel to get around the Bayer issue, you'd have the same number of pixels. Because of the AA filter, I doubt a half pixel shift would gain anything useful. Someone can correct me if I'm wrong about that.

Well, if you do a nice shifting pattern, an AA filter can be counter-productive, as it is no longer needed because shifting itself can be anti-aliasing.  That doesn't mean that shifting is useless for a sensor that already has an AA filter, though.  It still increases resolution and reduces aliasing, which happens easily enough in the red and blue channels despite an AA filter.

Of course, not all shift patterns are created equal, and some may leave significant artifacts and fail to approach a virtually analog image.  Shifting an AA-less sensor to 4 positions to create no increased pixel resolution but full color at each pixel, for example, still leaves significant potential for aliasing, as each resulting pixel is still non-overlapping with its neighbors; a box sample.

matejphoto wrote:

It would be great if they could do it.

Something like 80MP (in real resolution) equivalent from a tripod would be nice.

Final pixel resolution can only be integer multiples of 45MP (45 to defeat the CFA, 90, 135, 180, 225, 270, etc).  If you are thinking of "equivalent" as the number of details with significant contrast, a la DxO's "Perceptual MPs", then of course there will be a falloff in increased returns as you go higher, but I don't think everyone appreciates how valuable an over-sampled image is as far as further resampling is concerned; if you did, say, 16 positions, the pixel-level detail would be very soft, but if you downsampled that directly to you display resolution, you will get fewer resampling and aliasing artifacts than starting from a single 45MP original image. Don't forget, this would be mostly used with still scenery at base ISO, so if you combine 4 images, you get half the noise; 16 images, 1/4 the noise, etc, with the actual ISOs being a small fraction of that of each exposure.  So, at ISO 50, a 16-position shift would yield a true ISO of 50/16 = ISO 3.125.  You can sharpen the hell out of that without introducing much visible extra high-frequency noise that sharpening creates when noise is visible or barely subliminal in a single image.

Stephane SHG wrote:

It sounds good but it's not that useable in real life. I have this on my olympus cameras and barely use it...

Sure, but other people may be doing more still-life photography than you (or anyone shooting active subjects), in which case they can get much better and more flexible captures. Shifting can greatly reduce aliasing and noise (un-shifted stacking can only reduce noise).

The potential benefits are not dissimilar to what cameras in phones are using to bring 200 down to 12.5.

Do we not see pixel shift in those types of cameras due to the difficulty in moving such small distances?