Thursday 11 June 2015

Sony! α7R II + RX10 II + RX100 IV = 14bit Raw?

Yes, I'm going there. Why? Because it says as much in the specifications for the cameras on the Sony website. Screenshots from the respective pages are included below (click the images to enlarge) along with direct links to the originating page. This is for the Sony a7r mark 2, the RX10 mark 2 and the RX100 mark 4.

What you are looking for is 'Sony ARW 2.3'

α7R II (ILCE-7RM2)

RX10 II (DSC-RX10M2)

RX100 IV (DSC-RX100M4)

For those who don't know, ARW 2.3 (lossy delta-compression) is the same format used on the first generation of the a7 series. Now I'm going to presume that if there were an update to the RAW method implemented by Sony, it would result in a change in version number, that's what version numbers are for. Others have done a fair bit of investigation into this to figure out just what this means to the captured file.

For example, a user called Iilah Borg over at the DPReview forums posted the algorithm.
RawSpeed, easier to read than dcraw. Results between dcraw and RawSpeed are the same.
  1. // Process 32 pixels (16x2) per loop.
  2. for (uint32 x = 0; x < w - 30;) {
  3. bits.checkPos();
  4. int _max = bits.getBits(11);
  5. int _min = bits.getBits(11);
  6. int _imax = bits.getBits(4);
  7. int _imin = bits.getBits(4);
  8. int sh;
  9. for (sh = 0; sh < 4 && 0x80 << sh <= _max - _min; sh++);
  10. for (int i = 0; i < 16; i++) {
  11. int p;
  12. if (i == _imax) p = _max;
  13. else if (i == _imin) p = _min;
  14. else {
  15. p = (bits.getBits(7) << sh) + _min;
  16. if (p > 0x7ff)
  17. p = 0x7ff;
  18. }
  19. dest[x+i*2] = curve[p << 1];
  20. }
  21. x += x & 1 ? 31 : 1; // Skip to next 32 pixels
  22. }
Over at others have been looking into this:
As discussed in some detail in the review of the Sony A7R, the Sony A7 and A7R and other Sony cameras all use an 11 + 7-bit lossy compression scheme: 11-bit range with 7-bit delta values from 14-bit pipeline (8 bits per pixel in effect). Compare that to a Nikon 14-bit lossless image file. There is a difference, and I am certain it explains some of the not-so-great pixel quality I’ve seen in field images (in localized areas, not in totality), but it has been hard to pin down.
8 bits per pixel? 8? Now I know log profiles and whatnot can do interesting things with storing dynamic range in captured images, but 8 does not equal 14. I thought a little bit more investigation was in order, so I scooted off to the Sony Community forums and searched for 'lossy compression' on there. This post (from over a year ago) highlights someone's frustration with Sony's lossy compression approach.
Sony - Nikon was sued for their D600 issues.  Why not do the right thing and address the A7r and lossy compression?

You have a number of experts discussing the A7r shutter shock and Sony's lossy compression scheme and not yet a response from Sony.

Sony advertises that their cameras provide 14 bit color but the Sony compression algorithm never delivers more tha 8 bits of color depth according to file sizes.  This is deceptive and may amount to fraud.

Users want a camera that at the least provides an option for TRUE 14 bit color depth (even at the expense of frames per second) and a software update that eliminates shooting in the shutter range where shutter shock and blur are an issue.
You can find a more technical explanation of this over at the fredmiranda forum. Now after all of this, the Sony's announcement still rocks. All of the other specifications seem top notch and we are talking about a serious 35mm frame camera contender against Canon and Nikon offerings, (considering the lens compatibility, available video recording resolutions, connectivity options) but as always, until these cameras are being used in the wild and images / video start being uploaded, everything else (regarding AF speed/ tracking, colour science (color for my American friends), video quality etc is going to be pretty much just excited speculation.

If true about the lossy format, if true it seems strange, lossless gives the customer more editing power. Don't tell me these cameras can do all of this other stuff but not provide a lossless image.


DPReview has conducted an interview with Kimio Maki, Sony's 'Senior General Manager of the Digital Imaging Business Group'. The question of raw compression came up, his reply was non-committal.
 Right now it is the same, yes. We’re still working on it. In the future we may change the software but that’s not completed yet. We have consumers who require 14-bit etc., and we’re considering [how to deal with it]. 
Doesn't really make sense. They are working on it, so in the future they may change it but it isn't completed yet. Huh? They either will implement or they won't, what's with this weird middle of the road if-but-maybe stuff? And it isn't just customers who require it, the competition already provides it. Seriously, the Alpha 7R II looks like a smashing camera, add a 14bit lossless raw already!

Meanwhile, back at the ranch...

Anyway, speaking of excitement I popped over to the Sony Camera YouTube Channel and had a brief look through the video uploads for these cameras. Now we are all probably quite used to these companies uploading underwhelming footage for their new products, but I noticed a few things in the videos, like quantisation pulsing during shots of the sea, check the wave motion, distracting highlight clipping (I don't mean that it is clipped, but how the clipped pixels render), macroblocking on motion, things like that. Is this a consequence of XAVC S (h.264)? Or is YouTube's own compression?

Probably a bit of both, I would like to be able to download the original files Sony has (before being uploaded to YT) to see myself. If you find a link, drop me a line, so I can waste more of my time pixel peeping and procrastinating from this video edit I'm supposed to be working on.

Still, congratulations to Sony, competition is good! :-)

Now if you don't mind, I'm going to make a cup of tea and watch Canon and Nikon sweat a bit teehee.

I'll leave you with some screenies from the Sony videos. There are not meant to be conclusive proof of anything or some snipe at these new products, merely stuff I randomly noticed while watching the videos. (Note the resolution of the video in the bottom right corner, if you're interested).

One more thing, BSI CMOS Full frame! Hubba hubba.

Slowing green bubble sticks of doom

Picasso cubism 
Sensitive to red (hopefully you can pull this channel down, something similar happens on my Samsung NX1)

High? Bye lights.

High entropy scenes are a great way to test compression algorithms

Clipping on that little bridge behind the tree