Those 14 bits

Started Jan 17, 2014 | Discussions
Shop cameras & lenses ▾
stoppingdown
stoppingdown OP Regular Member • Posts: 405
Re: NEX-7 is 14-bit

spacemn wrote:

Again bottom-line is still that no one - several hundred pages into the discussion - has come up with comparable Sony A7(r) and Nikon D600/800 RAW's that people can put through their paces in their favourite RAW convertor, where we can actually see if there is any apparent difference in the robustness/richness of the files.

Sigh. This is true, and that request was in my first post of this thread. I'm a strong believer of the approach "facts first, then discussion". I was going to write a "I give up" post, but some hope came from an exchange between tesilab and MisterHairy... Let's wait and see.

-- hide signature --

Fabrizio Giudici
http://stoppingdown.net

 stoppingdown's gear list:stoppingdown's gear list
Nikon D7000 Sony Alpha NEX-6 Sony a6000 Nikon AF-S Nikkor 300mm f/4D ED-IF Nikon AF-S Teleconverter TC-14E II +8 more
SQLGuy Veteran Member • Posts: 4,917
Re: NEX-7 is 14-bit

spacemn wrote:

Again bottom-line is still that no one - several hundred pages into the discussion - has come up with comparable Sony A7(r) and Nikon D600/800 RAW's that people can put through their paces in their favourite RAW convertor, where we can actually see if there is any apparent difference in the robustness/richness of the files.

Why do you say this? The analysis here: http://www.dpreview.com/forums/post/52908651 is based on A7R and D800E raw files from the same IR test scene. You can download those files from IR too.

-- hide signature --

A7 with kit lens and a number of legacy lenses (mostly Canon FD)

 SQLGuy's gear list:SQLGuy's gear list
Canon PowerShot G9 Nikon D200 Sony Alpha NEX-7 NEX-5T Sony Alpha 7 II +5 more
stoppingdown
stoppingdown OP Regular Member • Posts: 405
Re: NEX-7 is 14-bit

SQLGuy wrote:.

Why do you say this? The analysis here: http://www.dpreview.com/forums/post/52908651 is based on A7R and D800E raw files from the same IR test scene. You can download those files from IR too.

Sure, those diagrams show a difference in an engineering perspective, but since in the end most people are interested in a photographic evaluation rather than an engineering one, it's important to see a photographic sample in order to understand whether there's a difference in perceived quality.

-- hide signature --

Fabrizio Giudici
http://stoppingdown.net

 stoppingdown's gear list:stoppingdown's gear list
Nikon D7000 Sony Alpha NEX-6 Sony a6000 Nikon AF-S Nikkor 300mm f/4D ED-IF Nikon AF-S Teleconverter TC-14E II +8 more
Iliah Borg Forum Pro • Posts: 21,535
Re: NEX-7 is 14-bit

I just want to understand that you are ruling out any phenomenon related to RAW decoding bugs og other issues in the process of decoding and demosaicing the RAWs?

Decoding: try putting a bug into Sony raw decoder and see where it will take you.

Bugs in demosaicking are totally irrelevant to this discussion.

What issues did you experience with the Fuji x-trans RAWs,

Why would we discuss Fuji here? There are several other forums where it is on topic.

I am not denying the fact that the compression algorithm Sony uses is lossy. I am just not convinced Sony's compression has any important impact on IQ

The specs state 14 bits. So Sony themselves see some benefit in 14 bits

Again bottom-line is still that no one - several hundred pages into the discussion - has come up with comparable Sony A7(r) and Nikon D600/800 RAW's that people can put through their paces in their favourite RAW convertor, where we can actually see if there is any apparent difference in the robustness/richness of the files.

While my bottom line is - people buy things based on published specifications. Why not to make those accurate?

-- hide signature --
SQLGuy Veteran Member • Posts: 4,917
Re: NEX-7 is 14-bit

stoppingdown wrote:

SQLGuy wrote:.

Why do you say this? The analysis here: http://www.dpreview.com/forums/post/52908651 is based on A7R and D800E raw files from the same IR test scene. You can download those files from IR too.

Sure, those diagrams show a difference in an engineering perspective, but since in the end most people are interested in a photographic evaluation rather than an engineering one, it's important to see a photographic sample in order to understand whether there's a difference in perceived quality.

-- hide signature --

Fabrizio Giudici
http://stoppingdown.net

So, you've inspected those images? Processed them in various ways to see how they react to manipulations? And not seen any difference that would concern you?

Or you're asking me to do that for you?

-- hide signature --

A7 with kit lens and a number of legacy lenses (mostly Canon FD)

 SQLGuy's gear list:SQLGuy's gear list
Canon PowerShot G9 Nikon D200 Sony Alpha NEX-7 NEX-5T Sony Alpha 7 II +5 more
pgb
pgb Senior Member • Posts: 2,005
Re: Those 14 bits
1

viking79 wrote:

SQLGuy wrote:

the A7R Flawed thread you'll see where it's proven through raw file analysis that the A7 and A7R are, in fact, returning only 11 bits of data from the sensor. The 3 lowest order bits are dropped before any processing even starts. The result is about 1700 distinct values from a test image where the D800 returns about 9000.

Ah, so this is the lossy compression people are speaking of? Has someone taken identical photos and done a difference to show what data is actually lost?

Most people don't seem to complain about actual RAW quality from A7r, but is this maybe responsible for some of the weird artifacts, like color rings people are seeing?

Eric

Overall the A7 is only compressing with less than a 2:1 ratio which is visually `loss less'. In the video and cine industry compression rates of 6:1 are used without a noticeable difference.
Apple's least compressed video codec, Prores HQ is 6:1.

Digital broadcasting uses around 40:1 compression.

I loaded a 2.4 MB Bmp into photoshop and saved a jpg at quality 12. The jpg was 800 MB or 3:1 compression, which is more compressed than the Sony. So it's compressing less than the highest quality setting in PS.

Log encoding can also used to squeeze 14 bit sensor data into 12 bit files.

Sony should provide the uncompressed option as well for peak users. It might be a slight improvement in some specific shots.

-- hide signature --

Peter

 pgb's gear list:pgb's gear list
Canon EOS 40D Sigma 17-70mm F2.8-4 DC Macro OS HSM Canon EF 28mm f/2.8 +8 more
seachicken2000
seachicken2000 Veteran Member • Posts: 3,201
Re: NEX-7 is 14-bit

Iliah Borg wrote:

seachicken2000 wrote:

Iliah Borg wrote:

Delta encoding is a lossy compression.

Why is it lossy?

Because in a 16-byte block for 16 pixels they have 11 bits for minimum, 11 bits for maximum, 2 4-bit indices of the locations of minimum and maximum values in the block and 14 7-bit deltas from minimum. So, if you have the amplitude of less than 128, it is lossless. But if you have a sharp edge with larger amplitude it is lossy.

The details are interesting, thanks.

If I understand it correctly areas of flat tone will be encoded without error. Areas that contain large differences in amplitude will have errors in their values.

The errors in these values are random in nature, and so will manifest themselves as noise around contrasty features? If so is this noise going to be visible in real images?

-- hide signature --

A rose by any other name is still a chicken.

 seachicken2000's gear list:seachicken2000's gear list
Sony Cyber-shot DSC-R1 Fujifilm FinePix S2 Pro Sony Alpha NEX-5 Sony Alpha NEX-7 Sony Alpha 7R +21 more
pgb
pgb Senior Member • Posts: 2,005
Re: Those 14 bits

Correction, that should be

The jpg was 800 KB ...

 pgb's gear list:pgb's gear list
Canon EOS 40D Sigma 17-70mm F2.8-4 DC Macro OS HSM Canon EF 28mm f/2.8 +8 more
Iliah Borg Forum Pro • Posts: 21,535
Re: NEX-7 is 14-bit
1

If I understand it correctly areas of flat tone will be encoded without error.

Yes.

Areas that contain large differences in amplitude will have errors in their values.

Suppose you have a horizontal feature in shadows which projects to a strand of 32 pixel long on the sensor (camera in landscape orientation; saying in shadows to to exclude any artifacts introduced by the compression curve). For accurate gradation recording the change in one colour channel along that 32-pixel long strand should be no more than 128, but if it is more the precision in recording is not 1 level, but (max-min)/128 and rounded to the power of 2. Posterization.

-- hide signature --
FuzzyQball
FuzzyQball Senior Member • Posts: 1,616
Re: Those 14 bits

More like 99%

-- hide signature --

Glenn

 FuzzyQball's gear list:FuzzyQball's gear list
Sony Alpha NEX-5N Sony a6000 Sony E 18-55mm F3.5-5.6 OSS Sony E 50mm F1.8 OSS Sony E 55-210mm F4.5-6.3 OSS +1 more
FuzzyQball
FuzzyQball Senior Member • Posts: 1,616
Re: Those 14 bits

Here is another question.  Can you see that type of fine tone changes on your monitor, or in print?

-- hide signature --

Glenn

 FuzzyQball's gear list:FuzzyQball's gear list
Sony Alpha NEX-5N Sony a6000 Sony E 18-55mm F3.5-5.6 OSS Sony E 50mm F1.8 OSS Sony E 55-210mm F4.5-6.3 OSS +1 more
Midwest Forum Pro • Posts: 16,640
Re: Those 14 bits

stoppingdown wrote:

(*) Which brings back the utility of a comparison with another brand: shall we perhaps discover that those 14 bits are marketing hype also for others?

Canon DSLRs have all been 14- bit for a long time. I have never seen them refer to that fact n their advertising; I doubt it could be called 'marketing hype'.

secondclaw Regular Member • Posts: 247
Re: Those 14 bits
1

Here is another question.  Can you see that type of fine tone changes on your monitor, or in print?

-- hide signature --

Glenn

I can speak for my experience. I owned 5D2, nex-7, and A7R, and have used a wide gamut monitor for years. After one night of shooting I noticed extreme amounts of posterization visible when looking at the raw images from a7r in lr3. After attempts with Sony's image converter proved futile (same issue), I posted about my problem on dpr ... though the thread quickly deteriorated into usual juvenile insults, I learned something from it ... when I move the image to my non wide gamut ips monitor, the pisterization is unnoticeable. It leads me to conclude that whatever data is lost by Sony algorithm affects wide color spaces only.
I have never noticed this problem with nex7.

 secondclaw's gear list:secondclaw's gear list
Canon EOS 5D Mark II Sony Alpha NEX-7 Sony Alpha 7R Canon EF 70-200mm f/2.8L IS USM Canon TS-E 17mm f/4L +10 more
Midwest Forum Pro • Posts: 16,640
Re: Those 14 bits

viking79 wrote:

viking79 wrote:

My guess is neither Sony sensor would benefit from 14 bit, so they didn't use 14 bits. The A7 sensor is a bit noisy anyway, so maybe that limits it to 12 bits.

Eric

Just to add, I thought someone was stating A7/A7r used 12 bits, I see according to the spec sheet they use 14 bits. It is probable that they might need that at base ISO for their added dynamic range. However, my argument is more general. A sensor only needs a certain number of bits, and increasing that further is pointless.

Having 14 bits on say a NEX 7 would fall under the pointless category. If they come out with a 54 MP Full Frame camera with similar sensor technology as current, 14 bits would be pointless, but how could they market it unless it was at least the same as A7/A7r? If they marketed it as 12 bits people would think it performed worse when it didn't.

So are 14 bit ADC marketing only for A7/A7r? Probably not, since Pentax K-5 clearly benefits from 14 bit at ISO 80 (It is same pixel pitch and similar sensor design as Sony A7r) so it would be reasonable to assume that the A7r might also benefit from 14 bit at base ISO.

The A7 we really don't have a comparable APS-C sensor or any other full frame camera to compare it to which uses a similar sensor but at 12 bit. With detailed spec sheet for the sensor it would be easily calculated, but that is not available to us. Maybe Erik or GordonBGood has done calculations with some RAW data to see what it needs?

This is an old article but the principles still would apply. There is a reason for 14-bit, and not just for color gradation:

http://www.earthboundlight.com/phototips/nikon-d300-d3-14-bit-versus-12-bit.html

Astrophotographer 10 Veteran Member • Posts: 8,489
Re: 14 bits compressed with no loss of information

Just out of interest I might add all Astronomy dedicated CCD cameras are 16bit. In fact I usually process in a 32 bit environment to retain maximum ability to retain detail in highlights/shadow areas.

When you process an image that is 16bit RAW you need several iterations of levels and curves to bring up the data as it is not all across what 16bit can record.

However when you do there is plenty of data in the dim shadow areas and detail in the highlights that become much harder to blow out. Well depth comes into blowing out highlights but most sensors are antiblooming meaning the extra spill over once full flows away in the circuitry rather than spiiling into surrounding pixels.

But having processed both types of files extensively I would say 16bit files have more dynamic range and fine detail. I note medium format cameras use 16 bit and from comparison images with D800e there are fine tonal graduations that hold up better in the MF image. Like shiny areas of facial skin retain their colour and detail better.

It would be a fine gradation, tonal differences retained slightly better dynamic range type improvement.

Its only a matter of time before digital cameras are all 16bit. I imagine processor power has been the restraining factor of why they aren't already. It takes some processor grunt to handle the files and that would make your average camera slow and laggy which would be unnacceptable.

Greg.

 Astrophotographer 10's gear list:Astrophotographer 10's gear list
Sony Alpha 7R II Fujifilm X-T2 Sony FE 55mm F1.8 Fujifilm XF 50-140mm F2.8 Fujifilm 16-55mm F2.8R LM WR +7 more
GaryW Veteran Member • Posts: 8,416
Re: Those 14 bits

SQLGuy wrote:

Let's say you go out and buy a new Corvette Stingray, specced at 460HP. You were considering a Mustang or a Porsche Cayman as well, but the 460HP was one of your deciding factors. Then you find out that it only makes 362HP. Do you say, "Well, has anyone run 1/4 mile times and flexibility tests to see how much difference it makes?" Or do you get mad at GM for falsely advertising a specification that they know is a decision factor for buyers?

P.S. This is just a hypothetical situation. I am in no way implying that the new Corvette will make anything less than its rated horsepower.

It's not a crazy analogy, tho.  Every engine is unique and will put out a slightly different amount, so manufacturers often understate the value just to cover themselves.  But it doesn't matter if it's 300HP or 400HP, it's whether or not it performs as desired.  If a 300HP car goes faster (say, because it's lighter), then it doesn't matter that a slower car has 400HP.  Or something.

-- hide signature --

A7 with kit lens and a number of legacy lenses (mostly Canon FD)

-- hide signature --

Gary W.

 GaryW's gear list:GaryW's gear list
Sony E 16-50mm F3.5-5.6 PZ OSS Sony E PZ 18-105mm F4 G OSS Sony Cyber-shot DSC-V3 Sony Alpha DSLR-A100 Sony Alpha NEX-5 +8 more
GaryW Veteran Member • Posts: 8,416
Re: Those 14 bits

secondclaw wrote:

Here is another question. Can you see that type of fine tone changes on your monitor, or in print?

-- hide signature --

Glenn

I can speak for my experience. I owned 5D2, nex-7, and A7R, and have used a wide gamut monitor for years. After one night of shooting I noticed extreme amounts of posterization visible when looking at the raw images from a7r in lr3. After attempts with Sony's image converter proved futile (same issue), I posted about my problem on dpr ... though the thread quickly deteriorated into usual juvenile insults, I learned something from it ... when I move the image to my non wide gamut ips monitor, the pisterization is unnoticeable. It leads me to conclude that whatever data is lost by Sony algorithm affects wide color spaces only.
I have never noticed this problem with nex7.

The problem with using color spaces is calibrating everything. It may be that your monitor is not calibrated and/or your system needs a profile to match.  In addition, if you use other than sRGB, it is more complicated.

So, I'd question your setup before Sony, but hard to know.

-- hide signature --

Gary W.

 GaryW's gear list:GaryW's gear list
Sony E 16-50mm F3.5-5.6 PZ OSS Sony E PZ 18-105mm F4 G OSS Sony Cyber-shot DSC-V3 Sony Alpha DSLR-A100 Sony Alpha NEX-5 +8 more
GaryW Veteran Member • Posts: 8,416
Re: Those 14 bits

pgb wrote:

viking79 wrote:

Overall the A7 is only compressing with less than a 2:1 ratio which is visually `loss less'. In the video and cine industry compression rates of 6:1 are used without a noticeable difference.
Apple's least compressed video codec, Prores HQ is 6:1.

Digital broadcasting uses around 40:1 compression.

Video is a bit different in that you can get away with less quality per frame as multiple frames will have complementary details, if you get what I mean.

I loaded a 2.4 MB Bmp into photoshop and saved a jpg at quality 12. The jpg was 800 MB or 3:1 compression, which is more compressed than the Sony. So it's compressing less than the highest quality setting in PS.

But JPEG is designed with perceptual encoding in mind. A simpler algorithm could arrive at the same file size but not look as good. I'm not sure where Sony's algorithm falls.

Log encoding can also used to squeeze 14 bit sensor data into 12 bit files.

Yeah, a non-linear scheme could preserve the overall DR.

Maybe if we had the algorithm in front of us we could just see what they did. Someone mentioned DCRaw.

Sony should provide the uncompressed option as well for peak users. It might be a slight improvement in some specific shots.

-- hide signature --

Peter

-- hide signature --

Gary W.

 GaryW's gear list:GaryW's gear list
Sony E 16-50mm F3.5-5.6 PZ OSS Sony E PZ 18-105mm F4 G OSS Sony Cyber-shot DSC-V3 Sony Alpha DSLR-A100 Sony Alpha NEX-5 +8 more
Iliah Borg Forum Pro • Posts: 21,535
Re: Those 14 bits
1

Maybe if we had the algorithm in front of us

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. }
-- hide signature --
secondclaw Regular Member • Posts: 247
Re: Those 14 bits

You're right its hard to know. Both of my monitors are Spyder3 calibrated, and I never observed this problem - with 5D2 and Nex-7. I don't preclude a mix of bugs in decoding software, use of certain color spaces, etc ... but right now the simplest explanation is data possibly being lost that wide color spaces magnify. Since nothing else has changed in my setup.

GaryW wrote:

secondclaw wrote:

Here is another question. Can you see that type of fine tone changes on your monitor, or in print?

-- hide signature --

Glenn

I can speak for my experience. I owned 5D2, nex-7, and A7R, and have used a wide gamut monitor for years. After one night of shooting I noticed extreme amounts of posterization visible when looking at the raw images from a7r in lr3. After attempts with Sony's image converter proved futile (same issue), I posted about my problem on dpr ... though the thread quickly deteriorated into usual juvenile insults, I learned something from it ... when I move the image to my non wide gamut ips monitor, the pisterization is unnoticeable. It leads me to conclude that whatever data is lost by Sony algorithm affects wide color spaces only.
I have never noticed this problem with nex7.

The problem with using color spaces is calibrating everything. It may be that your monitor is not calibrated and/or your system needs a profile to match. In addition, if you use other than sRGB, it is more complicated.

So, I'd question your setup before Sony, but hard to know.

-- hide signature --

Gary W.

 secondclaw's gear list:secondclaw's gear list
Canon EOS 5D Mark II Sony Alpha NEX-7 Sony Alpha 7R Canon EF 70-200mm f/2.8L IS USM Canon TS-E 17mm f/4L +10 more
Keyboard shortcuts:
FForum MMy threads