Olympus E-M5 II Image Samples for Multishot (Pixel Shift)

[Guillermo Luijk]

Re-thinking about the new multishot mode, which blends 8 shots into a single output RAW in a rather complicated way prone to artifacts when moving subjects appear on the scene, I really wonder now why Olympus (or any other camera maker) doesn't implement a genuine HDR mode, i.e. multishot with diffent exposure levels ending in a single output RAW file with enhanced DR.

I thought this had never been done because camera makers didn't want to take risks in producing RAW files containing artifacts that could provoque criticism from users who didn't understand the limitations of the multishot concept. With this enhanced resolution multishot they make it clear this was not the problem.

ZERO NOISE VIRTUAL RAW

Regards

[Bart_van_der_Wolf]

Re-thinking about the new multishot mode, which blends 8 shots into a single output RAW in a rather complicated way prone to artifacts when moving subjects appear on the scene, I really wonder now why Olympus (or any other camera maker) doesn't implement a genuine HDR mode, i.e. multishot with diffent exposure levels ending in a single output RAW file with enhanced DR.

Hi Guillermo,

I've also been wondering about that for years already. Even just using multiple read-outs (which are non-destructive on CMOS devices) from the same capture, would allow to average and reduce random (read)noise, and thereby improve the S/N ratio. Combining that with different exposures (with the risk of subject movement) would also allow to improve the shot noise performance. Averaging 4 read-outs would cut read noise in half, and each quadrupling of the exposure would cut shot noise in half.

Cheers,
Bart

[spidermike]

It would be of interest if Olympus will allow the export of any of the 8 individual 16 MP subsamples as a raw file.

I think you can export not only the 8 indivudal files but also a combined 'raw' file.

[BobDavid]

Shoot in multi-shot mode first, then shoot a single shot. In post, you can paint out the artifacts on the top layer mask to reveal the single shot image below. That is how to workaround artifact issues. Multi-shot photography isn't a big deal once you "get it."

[spidermike]

This is an interesting overview:

http://robinwong.blogspot.co.uk/2015/02/olympus-om-d-e-m5-mark-ii-review_10.html



Contrary to my earlier post "You do NOT get to keep all the indvidual images, you get either 40MP JPEG or 64MP RAW (or both), with a backup of first frame of 16MP RAW if high resolution shot is used. "
It is impossible (in practice) to use the 40MP for portraits - which makes me wonder in what circumstancs it will be useful in practice.

[SZRitter]

This is an interesting overview:

http://robinwong.blogspot.co.uk/2015/02/olympus-om-d-e-m5-mark-ii-review_10.html



Contrary to my earlier post "You do NOT get to keep all the indvidual images, you get either 40MP JPEG or 64MP RAW (or both), with a backup of first frame of 16MP RAW if high resolution shot is used. "
It is impossible (in practice) to use the 40MP for portraits - which makes me wonder in what circumstancs it will be useful in practice.

My ideas: Still Lifes (studio or other), Architecture, and landscapes with long shutter speeds. In all honesty, I see lots of use for the feature in the ways I currently shoot.

And an area that is a little less common, but one I have, duplicating negatives/slides.

[Guillermo Luijk]

This is what artifacts can look like in a moving seascape:



200% crop:


Picture by diegogm.es

Since artifacts always seem to follow a pattern in diagonal lines, I don't think it must be difficult to write a routine to automate the correction in those areas by using the 16Mpx shot upscaled to 64Mpx, or simply blur the image with 1 pixel radius.

Regards

[Neil Folberg]

Hello,
Is this shot with the new EM5II using multishot? Would you be willing to make a full resolution file available for download?
Many thanks, Neil

[capital]

This is what artifacts can look like in a moving seascape:

Thank you again for the links to these images, Guillermo.

In some ways this jagged effect reminds me of heavily compressed JPEG artifacts, of course the multi shot effect isn't due to the jpeg rendering.

[Guillermo Luijk]

Hello,
Is this shot with the new EM5II using multishot? Would you be willing to make a full resolution file available for download?
Many thanks, Neil

It is the 40Mpx JPEG

http://dl.dropboxusercontent.com/u/12069587/Foros/_1140051.JPG

[SZRitter]

It is the 40Mpx JPEG

http://dl.dropboxusercontent.com/u/12069587/Foros/_1140051.JPG

Thanks for the link and the test shot!

So, I did a quick bit of experimenting, and since the vertical lines appear to be about one pixel, you can downsize it by 50% and the artifacts completely disappear. And with how high a quality that 50% is, you could probably print it very large. So, just think of the 40MP as super-sampling and downsize, and you still have some incredible imagery.

Just my 2 cents though.

[SZRitter]

So my math is a little funny, 50% downsizing creates a 10MP image. Hmmm..... maybe not as impressive as I thought. That said, I am also having good luck with another technique on your image. But, it isn't without it's drawbacks, and a very manual process.

[Guillermo Luijk]

Replacing affected areas shouldn't de an issue since they are moving parts, so a blurred version should suffice.

The interesting part is to automate the detection of the zones to be fixed. On individual RAW data (i.e. firmware) this should be quite easy just by detecting changes. On the blended file (RAW or JPEG) other strategies should apply to correlate with the artifacts.

Regards

[BJL]

http://robinwong.blogspot.co.uk/2015/02/olympus-om-d-e-m5-mark-ii-review_10.html
"...you get either 40MP JPEG or 64MP RAW (or both), with a backup of first frame of 16MP RAW if high resolution shot is used. "

Robin Wong implies that the 16MP raw that you get along with the 64MP raw and 40MP JPEG files in "raw+JPEG" mode is simply the first frame.  That's good news if true, since one can then have the option of using a special "motion effect healing brush" to use that first frame data in parts of the image that are Affected adversely by motion, while keeping the higher resolution elsewhere.

I would still like someone to automate this, based in data analysis to detect motion artifacts!

[Bart_van_der_Wolf]

Robin Wong implies that the 16MP raw that you get along with the 64MP raw and 40MP JPEG files in "raw+JPEG" mode is simply the first frame.  That's good news if true, since one can then have the option of using a special "motion effect healing brush" to use that first frame data in parts of the image that are effected adversely by motion, while keeping the higher resolution elsewhere.

I would still like someone to automate this, based in data analysis to detect motion artifacts!

Wouldn't the difference between 'the upsampled version' and the higher resolution version show the highest modulation in the motion affected regions?

Whether one uses the smaller first exposure companion, or a down-sampled/upsampled (or just properly blurred) version of the full resolution file for the difference is another matter. Technically one could use a properly blurred version of the full resolution result as a difference file, and look for the highest difference amplitudes as most likely regions needing to be addressed.

If the pattern is mostly diagonal, one could make a custom filter to either amplify it or suppress it.

Cheers,
Bart

[SZRitter]

My understanding of a jpg is that it is just an array of pixel data. Hypothetically, you could loop through the array and look for patterns in that data (oversimplified version of the actual programming). When you found the matching patterns, you could average the pixel data to make that spot "blurred". That said, I could see patterns, say feathers on a close up of a bird, possibly screwing with this.

However, and maybe it will be baked into the converter for Photoshop, if you have the original images, you should be able to look for the anomalies, then see how they compare to the originals, and if the pattern doesn't exist in the originals, smooth it out in the combined images. Keeping all of these arrays in memory, however, will be very memory intensive.

Of course, this all involves math much beyond my skills.

[capital]

CameraLabs has a nice write up explaining the raw files associated with High Res. mode and has some sample photos on a variety of subjects in this mode. The camera outputs a 40mp JPEG because Olympus believes it is not a true 64 MP worth of image data. When you shoot in high res. raw mode the camera delivers the 64 MP raw and an .ORI 16 MP raw image based upon on the first position in the 8 frame shift sequence. Apparently Olympus is down playing this mode and its use cases, so I am unsure if they will natively support corrections needed for the shift aliasing.

http://www.cameralabs.com/reviews/Olympus_OMD_EM5_Mark_II/

[rdonson]

It appears this is just the first salvo from Oly.  Based on this interview I think we can expect further refinements in this hi-res approach.  I wonder if we'll see this system adopted for the OMD E-M1?

http://www.dpreview.com/articles/5476551037/interview-with-setsuya-kataoka-from-olympus-om-d-high-resolution-mode

[SZRitter]

It appears this is just the first salvo from Oly.  Based on this interview I think we can expect further refinements in this hi-res approach.  I wonder if we'll see this system adopted for the OMD E-M1?

http://www.dpreview.com/articles/5476551037/interview-with-setsuya-kataoka-from-olympus-om-d-high-resolution-mode

Probably. Rumor is that Pentax is also looking at it for their DSLRs.

[Guillermo Luijk]

Some thoughts from this multishot sample from focus-numerique.com:

• Each single shot from the bird is sharp so shutter speed was very quick
• The absence of rolling shutter artifacts makes think the electronic shutter on the E-M5 is also fast enough
• Putting all that together means the delay in re-positioning the sensor for each shot is the place Olympus must be looking for improvement right now to allow multishot handheld