40 MP - Inspired by Olympus

Discussion in 'Olympus Cameras' started by wjiang, Feb 5, 2015.

  1. wjiang

    wjiang Mu-43 Hall of Famer

    I got curious after all this news about 40 MP tripod shots... so I did a quick little test on a tripod.

    Here's a normal 16 MP E-M5 image with the PL25, f/4, ISO200, 1/20s:
    mini.
    The full size is here: https://dl.dropboxusercontent.com/u/38795710/mini.jpg

    Here's a 4x stitch of 4x mean stacks (16 images total) using the same settings, just different framing:
    mega.
    The full size is here: https://dl.dropboxusercontent.com/u/38795710/mega.jpg

    Sorry I didn't try too hard to get the framing right... I also didn't quite get the correct size after the stitch (it was only about 35 MP) so I up-res'd it in Photoshop to 40 MP.
    Curiously, notice the DoF appears to have gotten shallower even though it's shot using the same aperture?

    Here's a 200% view of the 16 MP shot to roughly match framing:
    mini-crop.

    Here's a 100% view of the stitched 35 MP up-res'd to 40 MP stack for comparison:
    mega-crop.

    And the 100% view of the 16 MP shot again for comparison:
    mini-crop.

    More resolution :thumbup:
    No moire :thumbup:

    Folks, if you don't mind taking a lot longer (and it probably won't be quite as good as what the Olympus does out of the box), you can try something similar too!
     
    • Like Like x 2
  2. lightmonkey

    lightmonkey Mu-43 Veteran

    480
    Dec 22, 2013
    nice but how would i do this trying to capture the reflection of alfred hitcock jumping across a puddle
     
  3. wjiang

    wjiang Mu-43 Hall of Famer

    Well, Alfred Hitchcock is dead so that makes it a bit difficult for a start.
    And besides, you couldn't do it with the high resolution mode of the E-M5 Mk2 either :tongue:
     
  4. Dave in Wales

    Dave in Wales Mu-43 All-Pro

    Nov 5, 2011
    West Wales
    200% of 16 MP, 100% of 35 MP.
    Are you comparing apples with oranges, I'd like to see 100% of 16 MP.
     
  5. wjiang

    wjiang Mu-43 Hall of Famer

    Sure, I'll upload a 100% crop as well as the full ones tomorrow. In the mean time you can get an idea by downsizing the 200% crop by 50%...
     
    • Like Like x 1
  6. Turbofrog

    Turbofrog Mu-43 Hall of Famer

    Mar 21, 2014
    How did you actually create this? Did you use a longer lens and stitch 4 quadrants together, or did you move closer with the same lens, or did you just move slightly left/right/up/down to try and emulate the sensor-shift drizzle and let Photoshop figure it out?
     
  7. dwkdnvr

    dwkdnvr Mu-43 Regular

    79
    Aug 8, 2012
    Denver
    I think it's pretty well understood that there are techniques available to do similar processing already - whether it be using a longer lens and stitching mosaics or else by using Photo Acute to stack multiple of the 'same' image. Having played around with Microsoft ICE for stitching, I had planned to start using the PL25 as my landscape lens since the stitching was rather painless and the results clearly crushed the 12-50@12mm (the 12mm prime would obviously be much better). The advantage of these post-processing approaches is likely that since they've been developed as general-purpose techniques, they may deal with minor motion within the sequence better. The disadvantage is obviously the extra work involved.

    I think there are 2 interesting elements to the in-camera approach that will be interesting to see how they play out in the real world.
    - the full-pixel portion of the image shift effectively captures full-color information for each photosite similar to a Foveon. This should in theory result in much more accurate color info than would be available through post-processing techinques.
    - The Imaging Resource review indicates that there are THREE files produced in hi-res mode: the 40Mp jpeg, the 63Mp raw, plus another 16Mp .ori file. This ori file is apparently something that will need a Photoshop plug-in to handle. This extra file might have 'original exposure' data which will allow more sophisticate motion-compensation algorithms to be applied. We'll have to see what this file actually does.
     
  8. wjiang

    wjiang Mu-43 Hall of Famer

    I moved closer. In practice if I actually did this seriously I would use a longer lens.

    I got 4 significantly different framings (top-left, top-right, bottom-left, bottom-right). At each framing I used high speed burst to get 4 exposures. At the pixel level these stacks are not going to be perfectly aligned, which is a good thing as the mean process will tend to reduce noise and such once they are aligned in Photoshop. In fact I noticed that the mean stack looked noticeably better than any single component frame of the stack. Finally I did an auto panorama stitch in Photoshop to combine the 4 stacks.
     
  9. fireblade

    fireblade Mu-43 Regular

    42
    Jun 19, 2012
    i believe this is the way the serious macro boys do their work.
     
  10. That is why the depth-of-fields are different. The same would also occur if you'd used a longer focal length lens from your original camera position.
     
  11. wjiang

    wjiang Mu-43 Hall of Famer

    I just updated the original post. Pixel peep away!
     
  12. pellicle

    pellicle Mu-43 All-Pro

    Feb 10, 2010
    Southport, OzTrailEYa
    pellicle
    an interesting experiment but quite different to what the sensor shift will do. For a start the sensor shift will use the same lens.

    I would be inclined to attempt small reframings (without moving the camera) by moving the tripod head up down left right by a "widge" (similar to a bees dick, but a metric equivalent) and see if you can assemble that using another algorithm.

    That is what the Oly system will be doing and you'll see a more accurate example (as your lens will then be the limiting factor)....

    http://photoacute.com/studio/
     
  13. wjiang

    wjiang Mu-43 Hall of Famer

    I'm not actually trying to replicate what Olympus is doing. For a start, I'm not controlling the movement precisely enough (and there's no way for me to do this without precise accelerometer measurements). I'm also not actually getting round the Bayer filter problem.

    As I said in the original post, I'm sure the Olympus way gives you much better results as it's an integrated processing pipeline. This is just an experiment to see what can be done with 'simple' technology.
     
  14. LowriderS10

    LowriderS10 Monkey with a camera.

    May 19, 2013
    Canada
  15. barry13

    barry13 Super Moderator; Photon Wrangler

    Mar 7, 2014
    Southern California
    Barry
    He moved it a lot closer, to get a fraction of the subject in each frame.
    So the DoF was affected by that as well (if not primarily).

    Barry
     
    • Like Like x 1
  16. dhazeghi

    dhazeghi Mu-43 Hall of Famer

    Aug 6, 2010
    San Jose, CA
    Dara
    I'm not sure about that. It all depends on how cleverly they are collecting and combining the data from the different shots. Any 2x2 stitch with a proper tripod and stationary subject gives you almost a 2x increase in linear resolution (or 4x overall). But combining images that have been shifted slightly to infer in-between pixels is a lot harder, and judging by DPR's tests, the per-pixel quality is not quite at the same level as a good stitch.

    Of course the flip side is that the process in the E-M5 II is far easier than stitching. You don't need to manually shift the view, shooting only one piece at a time, and combining them later.
     
  17. OzRay

    OzRay Mu-43 Legend

    Jan 29, 2010
    South Gippsland, Australia
    Ray, not Oz
    One thing to note is that, at the moment, Olympus has chosen to offer a 40MP image, but what if they offered different resolutions, say 24MP and be able to do so with less than static subjects? Once you can do it a high MP count, why not a lesser count?
     
  18. KBeezie

    KBeezie Mu-43 Top Veteran

    697
    Sep 15, 2012
    Grand Rapids, Mi
    Karl Blessing
    Though they would probably utilize a rigging so that when they 'shift' the camera in any direction it stays fixed along that film/sensor plane. Sort of like how some of the copy-work rigs were set up for flat reproduction and archiving (though the oldest ones didn't shift at all, since it was usually large format cameras and they had the shift/etc built into the front with the lens board).

    In terms of macro work, They would have to make sure it doesn't move from that angle since a tilt/etc even a millimeter is going to vastly alter the perspective of the object.

    But yea, definitely something I imagine some people did in a digital workflow for high quality reproduction/archiving, if they weren't already just using a scan back on a large format camera, and shifting the back around the plane.