Details on small raw

Discussion in 'Nikon' started by fluppeteer, May 22, 2014.

  1. Since it's come up recently in discussions on the D4s and (optimistically) the firmware updates for the D800, and at the risk of re-posting from a rumour site, I thought some forum members might be interested in a rawdigger analysis of the sRAW format that Nikon's using.

    I, and many others, have been hoping for a decent "small raw" implementation for some time, especially with the D800 - there are many times when having the ability to handle a wide dynamic range nicely in post-processing would be useful even if a 36MP image isn't. Canon have had a "small raw" format for a long time, although there has been plenty of criticism about how useful it is.

    Nikon, to summarise rawdigger's findings, seem to have done a Y'CbCr 4:2:2 conversion, including a gamma correction step, using slightly under 12bpp. They point out that the file size, being downsampled but uncompressed, is not significantly smaller than using 12-bit lossy compression on a full resolution file - which correlates with Shun's findings that the "small raw" isn't particularly "small". (I guess it's appreciably smaller than an uncompressed full-resolution file, but since I always use lossless compression, I wouldn't expect massive savings, given that I'd be giving up 14-bit support.) They're a little concerned about how the representation affects the ability to do significant colour and tonal adjustments, which is, after all, the point of having a raw file.

    I still think Nikon's small raw is a work-in-progress (though from the above, it sounds like an early-out from the JPEG engine, if anything); that the D4s can't post-process it in the camera indicated to me that support is still being developed. Presumably applying lossless - or even lossy - compression to the small raw file actually would provide some useful size savings, but that's not yet an option.

    One of the (many) rumours floating around is that Nikon might release a D800 refresh at some point, one significant update of which would be the addition of a small raw (it sounds like an Expeed 3 to Expeed 4 update, similar to the D4 to D4s refresh). If that rumour turns out to be true (and, obviously, this isn't the place to discuss what the probability of that might be), this small raw implementation isn't going to convince me that I have an urgent need to upgrade.

    If the sensor output is linear, I'm actually not sure what the problem would have been with just averaging quads of colour values, especially if raw converters need changing anyway (which they do, for this Y'CbCr approach), in order to weight their interpolation properly by the adjusted centre positions. (You could possibly do some processing to allow for this in the raw file - I've not done the maths - but doing so might also end up throwing away data, so it may be detrimental.) Phase One seem to manage this with their "Sensor+" approach - although from the description I've found, I really hope they've not managed to get a patent just on binning a diamond arrangement of green sensels. (The coverage of a green pixel in a Bayer pattern is obviously a square that's rotated 45 degrees, so I'd argue that this is the obvious solution, and I'd expect some prior art - but this isn't the first time I've made that argument about a patent...)

    Anyway. to quote Bop-It, Nikon: "do it again, but better."

    (But I'm grateful for the recent D800 BIOS updates, so I don't mind Nikon's firmware team having a few days off first. I must drop some chocolate off at Nikon UK HQ. No harm in bribery...)
  2. The file size is insignificantly smaller than full resolution 12-bit lossy compressed NEF​
    So what's the point? There's no space savings to be had - and it remains to be seen if sNEF has any advantage over shooting NEF lossy compressed. Apparently, two advantages of shooting RAW - adjusting white balance in post and recovering highlight detail - are compromised in sNEF. All what's left then is the reduction in pixel count - which might speed up post-processing a little. What am I missing?
  3. Dieter: Quite. And that seems to be Shun's conclusion as well. Which is unfortunate, because the D800 really could do with a binning mode. (There's always DX crop, but that throws away sensitivity - although not necessarily more than the small raw processing does.)

    Like I said, it's not - from the reports I've heard so far (and, while I've handled a D4s, I've not tried small raw) - quite what we were all hoping for, sadly. It's possible that some patents are in the way of doing this better, but I'm guessing there, and my fundamental hatred of the number of obvious things that get patented might be affecting my cynicism levels.
  4. ShunCheung

    ShunCheung Administrator

    People's demands are very weird. 6, 7 years ago when Nikon DSLRs were mainly 12MP, e.g. D3 and D700, they were talking about Canon's "huge advantage" with 21MP DSLRs such as the 1Ds Mark III and 5D Mark II. Now Nikon DSLRs are mainly 24 and 36MP, they want small RAW to reduce file size.
    I, for one, don't like Nikon's implementation of RAW small on the D4S. A regular D4S image is 16MP: 4928x3280 while RAW small is half the dimension on both sides: 2464x1640, i.e. about a 4MP image, which has limited applications. However, the file size is only a bit smaller than the 14-bit lossless compressed full RAW I usually capture. This thread on DPReview has more comparison:
    Worse yet, on the D4S, full RAW and RAW small is an either/or selection. If you shoot RAW, you have to choose one. You cannot shoot RAW + RAW small as you can with RAW + JPEG. I think it is a very bad idea to permanently throw away information at the time of capture. Given that memory card and disk drive prices are so cheap in these days, I would avoid RAW small altogether. If you need a smaller image file for immediate display on some news web site, just shoot RAW + JPEG small and send the JPEG over.
  5. Shun: For the D4s, I'd completely agree with you. I never used the DX crop on the D700 for a similar reason - not enough pixels left over after shrinking. The D800 is a slightly special case - I absolutely want the option of 36MP, but the camera would be more flexible if I also had the option of creating a 9MP file (because 9MP is still - just - enough to be useful) if I wanted the resulting file size was significantly reduced. (The current small raw implementation may be a step towards enabling a compressed mode, as with other raw files.) If Nikon didn't have compressed raw, we'd all be grateful for the savings from the small raw mode. It's an odd fit for the D4s, but that may be a function of which camera got the Expeed 4 first.

    There are times when I know I'm not going to need (or get) 36MP. Either the result is for the web so I know I won't want the resolution or I believe I'm going to have enough noise/motion blur/lens softness that I don't think I'm getting anywhere near 36MP in the first place (DxO claims the 28-200 that I'm so fond of on the D700 is only capable of resolving 10MP on the D800E, and that's at the good end of the zoom). If the file size was reduced or the camera was sped up in any way, I'd sometimes be prepared to risk it, at least on the D800 - I've used the DX crop mode for the same reason, and I've certainly resorted to JPEGs for fast-moving sequences. So if it worked well, I'd want it, at least on the D800. For now, it appears not to work so well, while still requiring a fair bit of dedicated support in a raw converter. Fingers crossed for the next version.
  6. Hard disk space and camera card storage are getting cheaper all the time.
    If I didn't want everything the camera sees, in effect, why would I be shooting RAW anyhow?
    Large jpeg works for me when I don't "want it all" (cf. K. Hepburn), or when I'm shooting in near dark or furiously.
    I can't imagine when or why I would ever use small RAW, and wish Canon had given me something of more use to me. If you want it, why not?
  7. If I didn't want everything the camera sees, in effect, why would I be shooting RAW anyhow?​
    It's a dynamic range thing. I may not want all the resolution that the sensor can capture, but I may want all the bit depth that it can capture. If the lighting's awkward, especially if I don't have time to fiddle with manual white balance and D-lighting settings, I'd still want to be able to recover shadows (and highlights) without them blocking unreasonably. JPEG, even if you ignore the compression artifacts, sometimes won't cut it. (If you can get the exposure right in camera, I agree that JPEG is perfectly reasonable - but I'm too used to tweaking in post, and I've had to do quite a bit of extreme recovery recently. If you're in extreme low light, I admit that my argument is less sound, because the sensor may have dropped to an effective 8bpp anyway.)

    But if Nikon's small raw format is throwing away a significant amount of the remaining dynamic range, it's removing a lot of the benefit of using it. It seems like a bit of an odd choice, though I realise these things are never as simple as they appear.
  8. Everyone wanted more and now that we have more, everyone wants less!
  9. There's no one camera that can do everything brilliantly. A D800 - or a D610 - does a lot of stuff very well. Being able to generate 9MP or (less convincingly, although I didn't feel too restricted by my 300D at the time) a 6MP image while still having all the advantages of raw offsets one of the disadvantages of the resolution increase - one that a lot of people have complained about with the D800. I do think that the decreasing cost of storage and increase in computer speed has made this less of an issue than people claim, but I'd still use this if it was available. If small raw was a significantly smaller file size, made the camera any faster, and retained the dynamic range of the input raw file, which doesn't seem to be the case with this format. Of course, if it did those things (depending how well...), the D800 would be nearer to being a passable substitute for a D3s, and I'm sure that Nikon would rather that people wanting a big buffer and high frame rate at a lower pixel count just bought a D4s.

    On a D4s, small raw seems somewhat less convincing, unless you know there's no way your image could be turned into a full page spread (4MP really starts to get to the "web only" range, and at least three of the screens in my house have that resolution or - sometimes significantly - higher). It's not going to make a significant performance difference, anyway. But I remain hopeful that it's a work in progress and an improved version may turn up.
  10. pge


    Of course, if it did those things (depending how well...), the D800 would be nearer to being a passable substitute for a D3s​
    ... and more like a true update of the D700.
  11. DSLR camera companies seem to give diddly-squat about dynamic range. The simple solution would be to apply a gamma curve at the analogue stage - i.e. use log amplifiers after the sensor and before the A/D stage. That way the ever-decreasing bit depth available for defining shadow detail could be almost completely circumvented. But it seems that analogue electronic skills are too Olde Skool to be considered by today's "if there ain't an off-the-shelf chip to do it, I ain't interested" so-called electronics designers. Same as imagining that brick wall colour filters with no bandwidth overlap are a good idea to use in a sensor.
    Funnily enough the human eye, with its almost 100% filter overlap and chemical/analogue gamma curve built in at the sensor stage, seems to work very nicely indeed.

Share This Page