Pages: [1]   Go Down

Author Topic: Pixel-Shifting Vs. Larger Sensors  (Read 141468 times)

Michael Erlewine

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1027
    • MacroStop.com
Pixel-Shifting Vs. Larger Sensors
« on: October 27, 2017, 08:54:41 am »

There are a number of threads here on the new Sony A7r3 mirrorless camera, and I even started one myself. It would be nice if we could keep this thread here on topic, since that is why I am posting it and perhaps post other issues on more general threads.

What I would like to discuss is why the A7R3 may be particularly useful to me and the reasons I feel this way. And, of course, I am trolling here for more information on this topic and other photographers with a similar bent.

The first “major” DSLR that I had was the Nikon D1X, sometime in 2001. And I have had almost all of the DSLRs from Nikon since then, at least of the landscape variety. Since I shoot close-up nature photos, I never cared about sports-related cameras, high ISOs, and autofocus.

Anyway, for me, there have been a string of cameras all the way up to Nikon’s recent release of the D850. In my case, it’s always been onward and upward, onward to more and better features and upward toward sensors with ever greater megapixels. And the last couple of years have been kind of a climax of sorts, at least a branching out of options. And of course, I was swept up in it all, especially the seeming-endless waiting, etc. I marched through buying (and returning) three medium-format cameras, a long time ago the Mamiya RZ67 (with eleven lenses) and more recently the Hasselblad X1D and the Fujifilm GFX.

And along in there I also bought and tested out the Pentax K3 and K1, mostly because of their pixel-shift technology. And I had the Sony A7S and A7R. I bought the A7R2, sold it, bought it again and sold it yesterday. I also ordered a copy of the A7R3 yesterday, mainly because of the pixel-shift feature, which brings me to my point in writing this.

Of course, like many of us I am in the habit of getting cameras with more and better pixels, and without really thinking about it I imagined I would like a 100 Mpx camera or even greater. However, I have been recently having doubts about this after getting the Nikon D850 camera, with its 45.7 Mpx.

I have a very big and fast PC, one with two GPUs, eight cores, a fast processor, 128 GB of RAM, etc. However, I did notice with the new Nikon D850, which has only a modest increase in megapixels, a difference in the computing power required. Keep in mind, that I stack focus, so I often have to process 100 or more large TIF files in the same batch. This takes time, and with the D850 it takes a little MORE time. Not that much, actually.

However, I can see that when we get 100 Mpx sensors, it will increasingly take more time (and storage). I keep all my stacked layers, so I have many hundreds of thousands of images by now. And this set me to thinking. 

Of course, I have wanted larger sensors, but not just for more megapixels, but for larger-sized photosites that collect more light. That is why I originally purchased a Sony A7s, for more light and larger photosites or whatever we call them.

By using the Pentax K3 and K1, both of which have pixel-shifting technology in them, I could see that they provided superior color and its resulting resolution, but I was not happy the way Pentax handled non-native lenses (of which I have a lot), so eventually it was more trouble than it was worth and the Pentax lenses did not make me happy. I like APO lenses.

So, my point and perhaps question here to those techsperts out there is: can we have a discussion here about perhaps not yearning for ever greater-sized sensors and concentrate more on improving the color and resolution in smaller-sized sensors, the ones we already are using.

I am happy with about 50 Mpx in sensor size, not less please, but perhaps I don’t need more. Since I don’t make prints of my images (never have), I only need a size to display on the web or place in an e-book format. Typically, I used images that are 2048 pixels on the long side for what I post, depending on where I post of course.

So, I’m wondering if my Nikon D850, which is great by the way, much nicer than I had imagined, along with the new Sony A7R3 (if it works as advertised) might be all that I need?  At 42 Mpx, the A7R3 is not much different than the 45.7 Mpx of the D850, and that may be as much as I need.

I am wondering, since I ONLY do still photography on a tripod, whether the pixel-shift technology of the A7R3 may give me the color (most important to me) and the enhanced resolution (however that works), so that instead of having to ever project myself forward to larger and larger sensors, I might (at least for a time) be happy with what I have (or will soon have with the A7R3)?

I am sure some of you here will have more technical thoughts about this conundrum I am in, either agreeing with me or pointing out something I have not thought of. Thanks for feedback.

P.S. This is the style of photography I tend to do, this with the D810 if I remember right.
Logged
MichaelErlewine.smugmug.com. Founder MacroStop.com, MichaelErlewine.com, YouTube.com/user/merlewine

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #1 on: October 27, 2017, 10:59:40 am »

I used to think that binning was the way to go. It's better to squeeze one bit of data out of many crappy sensels than it is to squeeze many bits out of one good sensel. What i didn't realise is just how much of an influence the undesirable lenscharacteristics would have when  the sensor behaves "perfectly".

This is quite evident if you spend some time with the current crop of pocket compacts with 20mpx. If you downsize for the web, you'll get perfectly smooth files, but even at poststampsize the lensdeficiencies are immediately obvious.

Which is why i eventually decided against an olympus with pixelshift technology. It would fit well for switching between thoughtful hi-res or lower res street and portrait, but the hi-res option requires the best of the best of lenses making the entire setup probably prohibitively expensive.

Of course, i don't know whether these thoughts help your situation where you also already own the best of the best lenses, but still... my 2cents worth.
Logged
Regards,
~ O ~
If you can stomach it: pictures

Michael Erlewine

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1027
    • MacroStop.com
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #2 on: October 27, 2017, 12:57:10 pm »

I used to think that binning was the way to go. It's better to squeeze one bit of data out of many crappy sensels than it is to squeeze many bits out of one good sensel. What i didn't realise is just how much of an influence the undesirable lenscharacteristics would have when  the sensor behaves "perfectly".

This is quite evident if you spend some time with the current crop of pocket compacts with 20mpx. If you downsize for the web, you'll get perfectly smooth files, but even at poststampsize the lensdeficiencies are immediately obvious.

Which is why i eventually decided against an olympus with pixelshift technology. It would fit well for switching between thoughtful hi-res or lower res street and portrait, but the hi-res option requires the best of the best of lenses making the entire setup probably prohibitively expensive.

Of course, i don't know whether these thoughts help your situation where you also already own the best of the best lenses, but still... my 2cents worth.

Thanks for posting, but I have the good lenses, so that is not the problem. Also, from many thousands of photos with the K3 and K1, I can see the improvement of this technique.
Logged
MichaelErlewine.smugmug.com. Founder MacroStop.com, MichaelErlewine.com, YouTube.com/user/merlewine

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #3 on: October 27, 2017, 01:17:14 pm »

Thanks for posting, but I have the good lenses, so that is not the problem. Also, from many thousands of photos with the K3 and K1, I can see the improvement of this technique.

I was wondering though, since you are already stacking so many files, with refocussing in between, shouldn't that provide enough jitter to overcome any coloradvantages that pixelshifting might give you?
Logged
Regards,
~ O ~
If you can stomach it: pictures

Michael Erlewine

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1027
    • MacroStop.com
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #4 on: October 27, 2017, 01:22:19 pm »

I was wondering though, since you are already stacking so many files, with refocussing in between, shouldn't that provide enough jitter to overcome any coloradvantages that pixelshifting might give you?

Not unless you "jitter" it. I don't do that or try not to do that. That's part of the process of focus stacking.
Logged
MichaelErlewine.smugmug.com. Founder MacroStop.com, MichaelErlewine.com, YouTube.com/user/merlewine

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #5 on: October 28, 2017, 07:59:29 am »

Hi,

If you shoot a red flower, only 1/4-th of the pixels will be used. So your 42MP Sony A7rIII turns into a 10.5 MP device. You shoot a green flower, you are using half the pixels. With pixel shift you will utilize all pixels regardless of colour and you will get very few colour artefacts.

Best regards
Erik
Logged
Erik Kaffehr
 

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #6 on: October 28, 2017, 09:00:00 am »

If you shoot a red flower, only 1/4-th of the pixels will be used. So your 42MP Sony A7rIII turns into a 10.5 MP device. You shoot a green flower, you are using half the pixels.

NO! Read the discussion about the Phase One Trichromat.

In addition, you can not just stack 4 RAW captures to yield true color at each sensellocation. The result would still show raster artifacts for even the slightest uniformity issues with lighting, sensor stability, and other inconsistencies between shots. It's an extremely sensitive process unfortunately. Some form of demosaicing or patternnoise reduction is likely necessary.
Logged
Regards,
~ O ~
If you can stomach it: pictures

Michael Erlewine

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1027
    • MacroStop.com
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #7 on: October 28, 2017, 09:38:32 am »

If you have not tried pixel-shifting of the Pentax K1/K3 type, then you have perhaps not seen the result, although there are posted shots all over the web. Having taken a couple of thousand pixel-shift shots, I liked what I saw, but the K1 was too finicky with non-native lenses, so I unloaded it. Pixel-shifting is a solid technique, with solid results, albeit on tripod and for still photos, which is what I do.
Logged
MichaelErlewine.smugmug.com. Founder MacroStop.com, MichaelErlewine.com, YouTube.com/user/merlewine

shadowblade

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2839
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #8 on: October 28, 2017, 09:51:51 am »

If you have not tried pixel-shifting of the Pentax K1/K3 type, then you have perhaps not seen the result, although there are posted shots all over the web. Having taken a couple of thousand pixel-shift shots, I liked what I saw, but the K1 was too finicky with non-native lenses, so I unloaded it. Pixel-shifting is a solid technique, with solid results, albeit on tripod and for still photos, which is what I do.

Any luck with it on landscapes? These are nonmoving from a no-AF-needed, unlimited-time-to-compose point of view, but often have subtle moving elements such as grass, leaves and running water. Can the software handle any degree of motion without creating artifacts?
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #9 on: October 28, 2017, 10:07:22 am »

Hi,

Things often work out better than expected. If you are shooting in studio, things are pretty stable. In the field, things that should not work often do. Just see electronic shutter on the IQ3100MP.

Regarding the Phase One Trichromatic, at this time I would say that information is extremely bad. Just a lot of claims P1 is making on a lot of fake information they call "science". Yes, it is that bad!

Best regards
Erik

NO! Read the discussion about the Phase One Trichromat.

In addition, you can not just stack 4 RAW captures to yield true color at each sensellocation. The result would still show raster artifacts for even the slightest uniformity issues with lighting, sensor stability, and other inconsistencies between shots. It's an extremely sensitive process unfortunately. Some form of demosaicing or patternnoise reduction is likely necessary.
« Last Edit: October 28, 2017, 10:10:49 am by ErikKaffehr »
Logged
Erik Kaffehr
 

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #10 on: October 28, 2017, 11:21:15 am »

Regarding the Phase One Trichromatic, at this time I would say that information is extremely bad. Just a lot of claims P1 is making on a lot of fake information they call "science". Yes, it is that bad!

True, but the discussion in that thread, although overly technical, has some useful info regarding overlap of colorfilters. Important take-away:
- colorfilters on the bayersensor always have an overlap in their respective sensitivity to the colorspectrum.
- overlap is necessary to discern the non-primary colors.

The hard nut to crack though is "how much overlap is optimal?".
- more overlap means less bitdepth for primary colors
- less overlap means less bitdepth for secondary colors.

More or less...
Logged
Regards,
~ O ~
If you can stomach it: pictures

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #11 on: October 28, 2017, 11:29:18 am »


I have a very big and fast PC, one with two GPUs, eight cores, a fast processor, 128 GB of RAM, etc. However, I did notice with the new Nikon D850, which has only a modest increase in megapixels, a difference in the computing power required. Keep in mind, that I stack focus, so I often have to process 100 or more large TIF files in the same batch. This takes time, and with the D850 it takes a little MORE time. Not that much, actually.

Pixel shifting as implemented in the a7RIII gives you four times as many images to deal with. WRT computer power, that's the bad news. Here's the next bit of bad news: in spite of most all image-processing tasks being highly parallelizable, most image processing programs don't go to the trouble of doing that. I have a 24-core dual Xeon system with 256 GB of RAM. The only image processing programs that routinely use all the cores and all the RAM I let them have are AutoPano Giga and PTGui. That is a shame. I don't know what you're stacking with, but next time it's stacking away, bring up a resource monitor and see how much of your system it's using.

Here's the good news. The image processing software folks are waking up to the long-simmering trend of slow to no CPU cycle time reductions as the years go by. If we want to go a lot faster, it's got to be parallel processing that gets us there. So the code is being slowly rewritten.

When you put GPUs into the mix, the story is different, but not all that different.

So help may be on the way.

Jim

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #12 on: October 28, 2017, 11:42:46 am »


So, my point and perhaps question here to those techsperts out there is: can we have a discussion here about perhaps not yearning for ever greater-sized sensors and concentrate more on improving the color and resolution in smaller-sized sensors, the ones we already are using.

Non-Bayer, or de-Bayered, photography goes back a long way in the early 90's I used mono sensors and filter wheels for some work. They did fine, it everything held still. Then along came Imacon with their de-Bayering piezo V-series Hasselblad backs. You can still buy similar technology from Hassy today. That works well, too.

So, in a sense, the a7RIII is a know quantity. The only issue is the implementation. I'm assuming the a7RIII will use voice coil actuators that already make the IBIS work.

I have no reason to think that Sony's implementation will be a lot worse than the piezo motion. If it's not, we'll be in the same situation as with Hassy piezo sensors, but at a much more attractive price point.

What's not to like? Well, we'll have four times as much data to deal with.

Different issue:

Michael, I know you have excellent lenses. I can tell you that a 50 MP sensor cannot capture anywhere near all the detail that they can lay down.

Jim

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Back to the wheel?
« Reply #13 on: October 28, 2017, 11:47:14 am »

Non-Bayer, or de-Bayered, photography goes back a long way in the early 90's I used mono sensors and filter wheels for some work. They did fine, it everything held still.

Maybe we could think about a return to that approach. When you think about it, the idea of installing a CFA on the sensor at great cost and some loss to IQ, then using four-shots and small shifts to try to undo it seems a bit daft.

And think about the great color we could get with 4 or five filters in the wheel!

Jim

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #14 on: October 28, 2017, 11:47:55 am »

If you have not tried pixel-shifting of the Pentax K1/K3 type, then you have perhaps not seen the result, although there are posted shots all over the web. Having taken a couple of thousand pixel-shift shots, I liked what I saw, but the K1 was too finicky with non-native lenses, so I unloaded it. Pixel-shifting is a solid technique, with solid results, albeit on tripod and for still photos, which is what I do.

I've seen the results from the PEN F (which also does oversampling) and it does indeed provide great results and visible improvements.
Logged
Regards,
~ O ~
If you can stomach it: pictures

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Back to the wheel?
« Reply #15 on: October 28, 2017, 11:58:08 am »

Maybe we could think about a return to that approach. When you think about it, the idea of installing a CFA on the sensor at great cost and some loss to IQ, then using four-shots and small shifts to try to undo it seems a bit daft.

And think about the great color we could get with 4 or five filters in the wheel!

Jim

Or a prism slit shutter? Considering that the readout of a scanline has become somewhat faster than in those days...?
Logged
Regards,
~ O ~
If you can stomach it: pictures

eronald

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6642
    • My gallery on Instagram
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #16 on: October 28, 2017, 10:03:45 pm »

If Sony wanted to, they -or an enterprising hacker- could presumably program the voice-coils to do

- numerous multiple sub-pixel shifts.
- serious shifts
- tilts
- small z-axis displacements

etc.

All of the above provide an opportunity to enhance captures in various interesting ways.

Non-Bayer, or de-Bayered, photography goes back a long way in the early 90's I used mono sensors and filter wheels for some work. They did fine, it everything held still. Then along came Imacon with their de-Bayering piezo V-series Hasselblad backs. You can still buy similar technology from Hassy today. That works well, too.

So, in a sense, the a7RIII is a know quantity. The only issue is the implementation. I'm assuming the a7RIII will use voice coil actuators that already make the IBIS work.

I have no reason to think that Sony's implementation will be a lot worse than the piezo motion. If it's not, we'll be in the same situation as with Hassy piezo sensors, but at a much more attractive price point.

What's not to like? Well, we'll have four times as much data to deal with.

Different issue:

Michael, I know you have excellent lenses. I can tell you that a 50 MP sensor cannot capture anywhere near all the detail that they can lay down.

Jim
Logged
If you appreciate my blog posts help me by following on https://instagram.com/edmundronald

Michael Erlewine

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1027
    • MacroStop.com
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #17 on: October 29, 2017, 02:13:59 am »

Any luck with it on landscapes? These are nonmoving from a no-AF-needed, unlimited-time-to-compose point of view, but often have subtle moving elements such as grass, leaves and running water. Can the software handle any degree of motion without creating artifacts?

As you point out, not so good with landscapes. As for handling motion, probably not, but I do mostly still life and in the winter, indoors.

Logged
MichaelErlewine.smugmug.com. Founder MacroStop.com, MichaelErlewine.com, YouTube.com/user/merlewine

davidgp

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 758
    • davidgp fotografia
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #18 on: October 29, 2017, 06:36:31 am »

As you point out, not so good with landscapes. As for handling motion, probably not, but I do mostly still life and in the winter, indoors.


For this photography I think pixel shift will be great for you... the only problem I see is workflow:

- it is my understanding that you don’t have to much love for the A7r II, if it is for the camera itself... maybe you also hate the A7r III for the same reasons, if it is for the color, apart of the advantages of pixel shift, Sony said they improved the color response of the camera... no idea if it is just their JPEG rendering.
- you will need first to create the image, and since this is not done in camera, you will have to add that process using Sony software... let’s see how optimize it is...
- as commented by other people, you will have double size files in terms of megapixeles.


http://dgpfotografia.com

davidgp

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 758
    • davidgp fotografia
Re: Pixel-Shifting Vs. Larger Sensors
« Reply #19 on: October 29, 2017, 06:38:27 am »

Pixel shifting as implemented in the a7RIII gives you four times as many images to deal with. WRT computer power, that's the bad news. Here's the next bit of bad news: in spite of most all image-processing tasks being highly parallelizable, most image processing programs don't go to the trouble of doing that. I have a 24-core dual Xeon system with 256 GB of RAM. The only image processing programs that routinely use all the cores and all the RAM I let them have are AutoPano Giga and PTGui. That is a shame. I don't know what you're stacking with, but next time it's stacking away, bring up a resource monitor and see how much of your system it's using.

Here's the good news. The image processing software folks are waking up to the long-simmering trend of slow to no CPU cycle time reductions as the years go by. If we want to go a lot faster, it's got to be parallel processing that gets us there. So the code is being slowly rewritten.

When you put GPUs into the mix, the story is different, but not all that different.

So help may be on the way.

Jim

I was just going to comment this... probably is more an issue if the software not being able to use all the resources available...

This should improve in the future, each day new CPUs with more core are appearing... so software companies will have to learn to parallel programming if they want to be competitive...



http://dgpfotografia.com
Pages: [1]   Go Up