Are 400mp stills coming to the Canon EOS R5?

[InchMetric]

For those that marvel at the idea that one could get this kind of update in a camera you bought years ago, you now know why Tesla owners are always happy.

 

[EOS 4 Life]

I wonder how fast this will work or if I can take a photo of a living thing (and yea with flash would be nice too..)

A tree is a living thing so maybe if there is not much wind.

 

[justaCanonuser]

Another nice application of pixel shift would be stuffed birds . To be serious, that's definitely a nice addition for some motifs. Maybe, one could create also art images with moving people, cars or animals by deliberately using the quirks of this technology - a bit in the style of Jeff Bridges funny 2-portraits-in-1-shot images which he creates with the relatively slowly rotating lens of his vintage Widelux camera ((C) Jeff Bridges):

 

[entoman]

I wonder how fast this will work or if I can take a photo of a living thing (and yea with flash would be nice too..)

Presumably it will work with electronic shutter, which means at *up to* 20fps (with fast shutter speeds, and a fully charged battery). So for a very slow-moving subject such as @neuroanatomist's sloth, you might get away with a handheld shot IF you have extremely steady hands. Flash photography with ES is not possible with the R5.

 

[LDS]

BTW, if Canon also aims to VR and 3D - as outlined in their plans - highly detailed images is what you need if the the viewer can move and "get closer" inside a VR environment - to minimize the distance at which the resolution loss makes the image unusable. It could be introducing this in the R5 could be "prototyping" to understand better what to do in future cameras.

 

[justaCanonuser]

Canon why you wouldn't start with 1TP = 1 terapixel = 1000MP where each JPEG file would require 1GB of memory? Complete nonsense. This is just a response to Samsung smartphones tiny sensors with 200MP.

Well, for those who want high-resolution stills of static motifs, pixel shift makes much more sense than a 400 MP sensor. But I am with you, 200 MP smartphone images are just marketing crap. The tiny pixels on those small sensors are technically rubbish, their dynamic range is extremely limited, and to catch really sharp images on the pixel level they would require crazy high shutter speeds which would be in conflict with the exploding noise. They try to overcome those problems with heavy use of algorithms, but that doesn't always work well. I have a Samsung smartphone with "only" 64 MP and four cameras, but its image quality is worse than the images from the single 8 MP camera in the old S5 smartphone I had before.

 

[neuroanatomist]

Presumably it will work with electronic shutter, which means at *up to* 20fps (with fast shutter speeds, and a fully charged battery). So for a very slow-moving subject such as @neuroanatomist's sloth, you might get away with a handheld shot IF you have extremely steady hands. Flash photography with ES is not possible with the R5.

Very steady hands, since presumably IBIS won’t be helping stabilize the image.

 

[roby17269]

Yes, they say so, then maybe it is discovered the image was actually shot with an Hasselblad. Jokes aside, street billboards are designed to be seen at a large distance - the actual resolution doesn't need to be high. But there are large images that are designed to be seen at close distance.

No, the photos were taken with iPhones, that's true. What they "forget" to mention is that those photos were not printed on the billboard as-is... very expensive hardware and professional teams were used to enlarge them

 

[entoman]

Very steady hands, since presumably IBIS won’t be helping stabilize the image.

That's an interesting point. I doubt if IBIS stabilisation would work in conjunction with pixel-shift. I'd expect one or the other to be operational only. Currently, with RF lenses, OIS and IBIS work in conjunction, but I'd assume that when pixel-shift is selected, that only OIS would be operational for stabilisation purposes.

 

[usern4cr]

Four 1/2-pixel moves would be the minimum to keep the same 3:2 aspect ratio. Canon has chosen nine 1/3-pixel moves for a higher MP output.

I’ve previously posted that I suspect even the 2x2 array, though it yields four times the MP, does not deliver four times the resolution. The reason is that gapless microlenses already collect light from the full spatial area projected into each pixel. Many years ago, my Zeiss cameras would capture 2x2 or 3x3 sub-pixel arrays with their 1.3 MP CCD sensors that did not have gapless microlenses, and where only about 1/3 of the pixel surface was photosensitive. That means the resulting 5 and 12 MP images were real resolution increases.

IMO, a 2x2 sub-pixel sampling array with a modern sensor will have a significant proportion of empty resolution, and Canon’s 3x3 sub-pixel sampling array will be mostly empty resolution.

I look forward to seeing some detailed testing.


Given the requirement for a static subject, I’d say no.

A thoughtful insight, which I totally agree with. In a single image, there will always be overlap between each color-filtered pixel sensor even with ideal conditions (a rigid flat in-focus subject, rigid camera & sub-pixel moved sensor perfectly aligned to the subject, and a perfect lens). Also, each pixel element has some (appreciable) light from all colors in it with a peak for the desired color, which the de-Bayering software is pretty good at resolving. Each shifted image (if 1/2 pixel) has the chance to give better color information since each pixel would now have each RGB input (and green would have 2 shots of every pixel), but 'should' only increase the pixel color & resolution accuracy instead of increasing pixel count. A 1/3 pixel shift with 9 shots would have quite a lot of overlap between pixels, which ideal de-Bayering software could probably get some impressive results with. If everything was perfect then I think that this could give an impressive and useful higher resolution image.

But 'perfect' in all these areas is impossible. Given the best Canon lens and careful testing setup with a carefully constructed test image, it will be interesting to see what results are found. But that will probably be done soon after the firmware update is provided, and when screen shots of small patches of the test image (in center and 4 corners at a minimum) are shown to us then we will all know just how good (or not) it turned out to be. I hope that we are all wowed by the results, but I expect it to be maybe a quarter better compared to what a (future Canon) 400MP sensor would give, with most of it just wasted blurred interpolation. But even a quarter better is still better than nothing. But if we only get a jpg image, as others have said Canon mentioned, then we won't get the chance to use our own (non-Canon) raw post software, which will be a shame.

 

[[email protected]]

400 MP is an interesting number. Seems like the only way to get to that high a number would be to take advantage of the DPAF split pixels (R5 really has 90 MP) along with pixel shift and some serious processing. Interested to see if the processing is done in camera or only in DPP and if the result is only a JPEG or something with 14-16 bits per pixel. If it works well, this will make the 85mm f/1.2 a must-have for some folks as that is just about the only lens with enough acuity to pull that off. Effective pixel pitch is just under 2 microns which puts DLA at about f/1.6, so you need a lens that is stunningly sharp at f/1.8. Seems like a great plan to move a bunch of those golden 85mm f/1.2s .

I may be mistaken, but I believe that - depending on the algorithm used - the stacking process thwarts at least some of the diffraction effect. This was my impression when using the Panasonic S1r, shooting around 180mp, just looking at the results. It was surprisingly fast and intelligent regarding some forms of moving subject (like water in landscapes). Order of magnitude better than Sony's - which I shot at the same time. I was often using f/11 for landscape shots using pixel shift, goosing the resolution, and it certainly didn't look like it was falling apart due to diffraction issues.

My impression is that it is more like image stacking for reducing noise, where it "breaks" the normal calculation for figuring out the likely noise level.

I'm sure people who understand the math better than I do will chime to correct me (and then others chime in to contradict those people

 

[[email protected]]

It must be doing this in camera otherwise this is nothing new, and can already be done simply by taking a burst of images and stacking them in post. Astro photographers do it all the time. I do agree that the buffer is large enough though 400mp is only 9 images

After exhaustive testing of all the CFexpress cards out there (did a review over at camnostic.com), I find that the buffer for the R5 is effectively about 45 frames, but about 20 of those frames are actually processed through to the card while the real buffer probably stands around 25 frames. In other words, in the time it takes the camera to take the 45 frames (a little over 2 seconds in H+ mode with electronic shutter), it moves about 20 frames over to the CFexpress card at the average speed, which boosts the perceived buffer to close to 50 frames. So buffer isn't really an issue.

If there is any in-camera capability for processing, I very much expect that the feature will require single frame shooting mode. If there isn't in-camera capability, the "effective buffer" could squirrel away about 5fps max, if it is taking 9 subframes per frame. Of course, the 20fps maximum fps of electronic shutter would limit a 9-subframe pixel shift function to about 2 fps.

-tig

PS: When I briefly owned an R3 for a month or so, before deciding to pass on it in favor of my R5's, I found that the increased buffer was probably actually the very same capacity as the R5 buffer, but it eked out more frames very roughly proportional to the amount the R3 files were shrunk down due to the lower resolution.

 

[Mt Spokane Photography]

Very steady hands, since presumably IBIS won’t be helping stabilize the image.

Yes, I recall going over the patent for the IBIS carefully when it came out. An electric pulse must be needed to shift the sensor and hold it in place while capturing the image. Ibis moves the sensor to account for camera movement so it seems unlikely that it could be stabilizing but moving the sensor at a offset of 1/2 pixel or whatever. I wonder if the mechanics are precise enough to allow that. Maybe it does not matter how much the offset is, in that case, IBIS could be possible by modifying the IBIS motion to be offset.

 

[Michael Clark]

There is already noise-reduction "baked in" to RAW files, i.e. in the camera firmware. New more efficient firmware could further reduce noise without impacting detail sharpness. You only have to look at software like DeepPrime to see that RAW files can have noise virtually eliminated. It will take a few years before camera processors using AI are powerful enough to deal with it at that level, but there's certainly room for improvement.

The NR applied to raw files is not based on firmware and applied in the image processor, it's on-die NR determined by the architecture of the sensor's electronics. The NR applied later on is a totally different step in the imaging pipeline.

 

[Michael Clark]

Most commercial photographers who shoot static subjects like product, food, interiors , reproduction, with cameras firmly fixed to a tripod will 90% of the time also use flash.

I’m pretty sure this feature will not offer flash sync just like focus stacking in the R5.

Yes, but the types of strobe those folks use have longer durations and could conceivably benefit from pixel shift in one of two ways:

Take the frames so fast that they are all captured during the "fat" part of the strobe's discharge curve.

Or

Leverage the fixed camera and subject to take each frame after an interval long enough to allow the strobes to fully cycle.

 

[David - Sydney]

Firstly, consider the ratio of amateurs : professionals who buy the R5. It's widely accepted that the vast majority of purchasers of high-end gear are amateur enthusiasts. Neither of us know the exact figure, but my guess is that pros make up less than 2% of R5 buyers.

To have such an extreme view of 1 out of 50 users would need some justification.

As for user-selectable final resolutions, I think it will be a simple choice between 45MP and 200MP, and nothing inbetween. I also think there's a strong possibility that the output will only be JPEG/HEIF, as the R5 probably doesn't have enough processing power to output 200MP RAWs, and would need a huge buffer ( or a VERY long busy" time).

The R5 outputs 8k30 at 2600Mbps, 4k120 (as 4k30 files with no sound) at 1880Mbps.
Shooting 9 raw shots and combining later is one implementation option as the existing buffer/bandwidth would be up to task.
Who knows how long the Digic X would take an internally combined shot. I would prefer it from a workflow perspective and I would assume that most users of the feature are fairly static subjects so we aren't really constrained by fps needed for action shots.

 

[David - Sydney]

And you can do a nine shots in a row of lightning to add them up for a 400 MP pic?
If the R5 is that "fast as lightning", it's good for you if that feature comes to the R5.

TBH that wouldn't be my approach to take photos of lightning but if that works for you ...

Agreed. Lightning triggers have always been a pain for me so using long exposures gets more keepers... YMMV

 

[David - Sydney]

With slow-moving birds it's easy to get the eye-AF to lock on and track, even when the bird is quite distant, if the background is uncluttered. But when the background is fussy and close behind the subject, I'm not having much sucess.

Using the standard AF case, I was getting lots of success shooting puffins which are fast movers. Really surprised me. I am not a birder was new to the genre but very happy with my R5. I'll admit that most of the shots had clear backgrounds to that helped.
Triple back button focus?

 

[Michael Clark]

Really? You believe that? How much is "a lot"? Do you have data to back that up? I'd guess maybe 1%...

More than 1% of photography using ILCs is astro, and practically every one of those photos are taken tripod mounted.

A good number of landscape photographers use tripods most of the time.

Anyone shooting at 400mm or longer and selling their work is almost certainly using a tripod with a gimbal mount, or at the very least a monopod.

Almost all product photography is done from tripods. Ditto for art reproduction, school portraits (every student in every school getting their picture taken every school year is a LOT of photos), any Macro work using focus rails, etc.

Your perception that the vast portion of the users of ILCs use only the same methods and practices as you do seems to me to be awful myopic.

 

[David - Sydney]

The Canon website stated is was jpeg, so I don’t know why there is still discussion on it.

Surely HEIF output would give a substantially better output than jpeg. Why do all that work and be limited to the jpeg colour space.