Pages: 1 2 [3] 4   Go Down

Author Topic: Yet some DNG comments (from a raw software developer)  (Read 38775 times)

BrianVS

  • Full Member
  • ***
  • Offline Offline
  • Posts: 156
Re: Yet some DNG comments (from a raw software developer)
« Reply #40 on: August 26, 2015, 07:51:48 PM »

Most people have not shot digital cameras long enough to have a manufacturer disappear and their raw file formats no longer supported by modern software.

I keep a Win95B machine around, with 5.25" floppy drives, 3.5" floppy drives, SCSI, ZIP, JAZ, and Clik. 30 year old 5.25" Floppy disks are readable, had Landsat data on them. Moved to a newer machine now. I remember Tintypes of family members that died long before I was born in my parents box of old pictures. I wonder what our relatives get passed to them 100 years from now.

20+ years ago my wife asked my to write code to unpack TIFF 6.0 images that made use of the multiple image stored in one file feature. Photoshop 3.0 would only process the first image. Writing the code was easy as the file format was documented. That is the advantage of DNG and any other standard that documents the file format. Why manufacturers of all cameras simply do not publish their file format is stupid and vain. It's just not hard to figure out, they would just be saving some people a lot of time. I still use the software written in the late 1980s to reverse engineer file formats. It's like a visual hex dumper that you write code to process the data as you go through it. A million and one uses.
« Last Edit: August 26, 2015, 07:59:31 PM by BrianVS »
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11969
    • http://digitaldog.net/
Re: Yet some DNG comments (from a raw software developer)
« Reply #41 on: August 26, 2015, 07:54:23 PM »

Here is why one photographer no longer converts to DNG https://photographylife.com/why-i-no-longer-convert-raw-files-to-dng#more-116618
I'm not buying it  ::) There isn't one point in the piece that can't easily be dismissed. None of the advantages of the DNG format, arguably most using an Adobe product was mentioned.
But then this was supposed to be a technical discussion of the format.
Logged
Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
I'm out of here.

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #42 on: August 27, 2015, 03:34:17 AM »

One problem which is inbetween technical and political is that Adobe Camera Raw doesn't have that good reputation concerning color rendition, while say Capture One has a good reputation, at least in the medium format segment from which I have most experience. Many assume that this is because DCP is bad and ICC is good. However, I can assure you that the problem is not DCP, but how Adobe has chosen to design the look of their color profiles. Their tone curve also gives a special highlight rolloff behavior that has quite some effect on the look, which one may like or not like. Many of the others use a pure RGB curve in the highlight rolloff which gives another look (Capture One on of them). Personally I don't think any of them is ideal.

With the LookTable 3D LUT component in the DCP this can be worked around though, and you can create an entirely different look which I do in my own DCamProf project. It would have been smoother if the LUT was a bit more flexible as discussed earlier in this thread, but it's still possible to do most things.

If we think about long-term archiving ability, I really think the camera SSF should get in there. Without SSF the future raw converters will have to rely on the color conversion algorithms used in current software, like DCP. With SSF you have the key to do any color rendition, which doesn't need to use any of the current color vision models.
Logged

BrianVS

  • Full Member
  • ***
  • Offline Offline
  • Posts: 156
Re: Yet some DNG comments (from a raw software developer)
« Reply #43 on: August 27, 2015, 05:38:18 AM »

I looked up "SSF" and found it was a file format for storing GPS data- then saw that you referred to it as spectral response.

One of my first jobs (The Carter Years) was writing code to convert raw imagery into radiometrically calibrated data, first from film and later from a custom-made two-color IR sensor. Having the spectral response of the film and the sensor was required for this. These were measured and stored separately from the imagery. Today, most manufacturers hold the spectral response of their sensors as "proprietary". Kodak was the exception so I have the spectral response for the KAF-10500 and KAF-18500 used in the M8 and M9. Leica published the spectral response for the M Monochrom.

How many other cameras have published spectral response curves?

http://www.awaresystems.be/imaging/tiff/tifftags/privateifd/exif/spectralsensitivity.html

Not sure if the above tiff-ep tag is sufficient, or if it was meant to store a full spectral response curve.
« Last Edit: August 27, 2015, 05:56:25 AM by BrianVS »
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #44 on: August 27, 2015, 08:24:08 AM »

I looked up "SSF" and found it was a file format for storing GPS data- then saw that you referred to it as spectral response.

One of my first jobs (The Carter Years) was writing code to convert raw imagery into radiometrically calibrated data, first from film and later from a custom-made two-color IR sensor. Having the spectral response of the film and the sensor was required for this. These were measured and stored separately from the imagery. Today, most manufacturers hold the spectral response of their sensors as "proprietary". Kodak was the exception so I have the spectral response for the KAF-10500 and KAF-18500 used in the M8 and M9. Leica published the spectral response for the M Monochrom.

How many other cameras have published spectral response curves?

http://www.awaresystems.be/imaging/tiff/tifftags/privateifd/exif/spectralsensitivity.html

Not sure if the above tiff-ep tag is sufficient, or if it was meant to store a full spectral response curve.

I was not aware of that tiff tag, interesting. Could be worth looking into.

Yes, SSF, Spectral Sensitivity Functions, is a good search term if you want to look for scientific papers on the subject.

Spectral response is usually found in the spec sheets of sensors, but one needs to measure it on the camera as the IR-filter and possibly other aspects affects the curves greatly. With a spectrophotometer, integrating sphere and monochromator you can measure it, or use a quicker method like Image Engineering's camspecs product. This is $15k gear (unless you buy old stuff off Ebay and write some own code) so it's not for the private individual, but I would not be surprised if Adobe is already using this when profiling their cameras. So it would be nice if their DNG converter would embed this measurement.

(Of course the lens modulates the SSFs too, but it's not necessary to model that aspect to start with, SSF with a standard lens gets you very far while different lenses only make minor changes on top)

The spectral response cannot be hidden away, anyone with this gear can measure it, and all the camera manufacturers have it for sure so they can measure competitors cameras if they'd like (I don't think they do, I don't see any reason why they would), so there is no purpose whatsoever to have this "secret", as the only people that you hide it from is consumers that doesn't have the measurement gear.

My own profiling project DCamProf can generate profiles directly from SSFs, tried it and it works well.

You can find some SSFs on the net published via scientific papers, but it's not widely available as SSFs haven't yet become an established form to do camera profiling. In many ways camera color is very conservative, we're stuck using models from the 1990s and there seems to be little interest in developing new ideas. Not that SSF is actually new, but computer power today has made it feasible to use more data and computing intensive models.
Logged

sandymc

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 331
Re: Yet some DNG comments (from a raw software developer)
« Reply #45 on: August 27, 2015, 09:46:09 AM »

In many ways camera color is very conservative, we're stuck using models from the 1990s and there seems to be little interest in developing new ideas.

Not that I disagree about it being time that something new should be available, but I haven't seen any models that would in practice (emphasis on "in practice" as opposed to undeniable theoretical advantages that some models offer) give the majority of users better results than what they can get now........

Sandy
Logged

BartvanderWolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6941
Re: Yet some DNG comments (from a raw software developer)
« Reply #46 on: August 27, 2015, 11:41:14 AM »

My own profiling project DCamProf can generate profiles directly from SSFs, tried it and it works well.

You can find some SSFs on the net published via scientific papers, but it's not widely available as SSFs haven't yet become an established form to do camera profiling.

Hi Anders,

I haven't gotten around to doing it myself yet, it's somewhere on my Todo list (for later this year?), but I'm wondering if we can roll our own SSFs on a budget and get good enough results. Especially with the difficulty of shooting Color targets with reflective patches, it should be possible to do better optically/spectrally when using something like the "Star Analyser 100" and make an image of a light slit.

The light slit will be broken up in a nice rainbow of spectral colors by the diffraction grating, and when the Raw data (in linear gamma) is analyzed, a list of relative luminances by wavelength should become available (after some calibration, e.g. on the Fraunhofer lines of atmosphere filtered daylight). I know that the concept is easier in theory than executed in practice, but these seem challenges that can be solved for relatively little money.

I think that the SSF path is potentially much more reliable than shooting variable ColorCheckers with a lot of surface reflection issues.

Cheers,
Bart 
Logged
== If you do what you did, you'll get what you got. ==

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1996
Re: Yet some DNG comments (from a raw software developer)
« Reply #47 on: August 27, 2015, 01:18:39 PM »

get good enough results.
for me good enough (with Sony A7) was in the end shooting just passport with matte patches (measuring it & measuring the actual illumination) and approximating SSF/CMF using a matlab script provided by people from RIT... granted not as good as elaborate designs with monochromators, etc discussed here (and elsewhere) recently, but for a solder-challenged me that's a solution.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #48 on: August 27, 2015, 02:25:12 PM »

Not that I disagree about it being time that something new should be available, but I haven't seen any models that would in practice (emphasis on "in practice" as opposed to undeniable theoretical advantages that some models offer) give the majority of users better results than what they can get now........

Sandy

Yes I think you're right. Doing anything new the first attempt would probably provide worse results than the current highly mature methods. If I had an established commercial product with many users I would be very very very careful to change anything that already works well.

What I'm thinking of improvements in this case is more automation, and less hand-tuning of profiles. However some sell the converters very much because of their hand-tuned profiles, the medium format manufacturers is the prime example I think.

What I'd like to see is profiles without those subjective tunings, just really scene-referred colorimetric, and then you would use spatially varying color appearance models to actually render the color image from that. There's 10+ years of research done in the academic world of this already, but no broadly established raw converter provides a possibility to work this way. I can't say for sure that it would work well though, but if I had an infinite amount of time I'd probably do some work in that direction.

DCP is often said to be better than ICC because DCP is scene-referred and ICC is not, but in practice DCP is just as output-referred as ICC. All Adobe's profiles apply a curve, and afaik all have subjective tunings too to provide a pleasing color appearance with that curve applied. You can't really work in Lightroom with a linear colorimetric profile and get good color appearance for general purpose photography, as all the tools to apply contrast will distort color in one way or another.
« Last Edit: August 27, 2015, 02:29:05 PM by torger »
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11969
    • http://digitaldog.net/
Re: Yet some DNG comments (from a raw software developer)
« Reply #49 on: August 27, 2015, 02:30:42 PM »

If we think about long-term archiving ability, I really think the camera SSF should get in there. Without SSF the future raw converters will have to rely on the color conversion algorithms used in current software, like DCP.
This is an idea a pretty smart color geek named Eric Walowit has been proposing for years. If you don't know him or of him, ping me off list, I'll give you his email address.
He's on Linkedin too:
https://www.linkedin.com/pub/eric-walowit/4/4a0/793
Logged
Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
I'm out of here.

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #50 on: August 27, 2015, 02:35:30 PM »

Hi Anders,

I haven't gotten around to doing it myself yet, it's somewhere on my Todo list (for later this year?), but I'm wondering if we can roll our own SSFs on a budget and get good enough results. Especially with the difficulty of shooting Color targets with reflective patches, it should be possible to do better optically/spectrally when using something like the "Star Analyser 100" and make an image of a light slit.

The light slit will be broken up in a nice rainbow of spectral colors by the diffraction grating, and when the Raw data (in linear gamma) is analyzed, a list of relative luminances by wavelength should become available (after some calibration, e.g. on the Fraunhofer lines of atmosphere filtered daylight). I know that the concept is easier in theory than executed in practice, but these seem challenges that can be solved for relatively little money.

I think that the SSF path is potentially much more reliable than shooting variable ColorCheckers with a lot of surface reflection issues.

Yes it would be really nice with an SSF measurement method on a budget. I think the colorcheckers can work quite well though. The glare issues are very real though, I've seen that, it's extremely difficult to shoot glossy targets. When making general purpose profiles you may not want to correct lightness so much though, and if you make a 2.5D profile you don't need to measure that many dark patches, which makes it a bit simpler to do well.

One of many cool things with having SSFs though is that you can profile against a database of full spectrum measured colors, that is for example real human skin rather than a test target that tries to mimic it.
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1996
Re: Yet some DNG comments (from a raw software developer)
« Reply #51 on: August 27, 2015, 03:07:20 PM »

This is an idea a pretty smart color geek named Eric Walowit has been proposing for years.
and why 'd a smart geek mr. Knoll ignoring that for years ? when he can just add a tag to DNG specification (covering DCP profiles) at will ?
Logged

sandymc

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 331
Re: Yet some DNG comments (from a raw software developer)
« Reply #52 on: August 27, 2015, 03:31:41 PM »

What I'm thinking of improvements in this case is more automation, and less hand-tuning of profiles. However some sell the converters very much because of their hand-tuned profiles, the medium format manufacturers is the prime example I think.

Yes, I'd agree with that. Or put another way - right now it's probably possible to make a lot of improvements without changing the color model as such, just by getting better at the camera profiling process. However I think that here ugly commercial reality get in the way; I don't think you can make much money from camera profiling software, because profiles can be copied freely.

Right now, the most companies that commercially produce camera profiling software either build very expensive professional packages, or do so as an accessory to hardware - e.g., X-Rite.

The best thing that e.g., Adobe could do for the world of camera profiling, IMHO, is to add the capability for licensing profiles to its software. That way software vendors could actually make money from selling software to build profiles, and we would see a lot of innovation.

Sandy
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1996
Re: Yet some DNG comments (from a raw software developer)
« Reply #53 on: August 27, 2015, 04:02:12 PM »

The best thing that e.g., Adobe could do for the world of camera profiling, IMHO, is to add the capability for licensing profiles to its software. That way software vendors could actually make money from selling software to build profiles, and we would see a lot of innovation.
people do buy (or pay subscriptions) Adobe products for totally different reasons nowadays... so Adobe in fact better served being a good samaritan and making it available for free.... yes.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #54 on: August 27, 2015, 04:20:34 PM »

The best thing that e.g., Adobe could do for the world of camera profiling, IMHO, is to add the capability for licensing profiles to its software. That way software vendors could actually make money from selling software to build profiles, and we would see a lot of innovation.

I'm not sure I understand. If profiles could be sold more easily than today, there would still be only a handful of profile producers, so the world market of profiling software would be like a handful :). Maybe we would see some inhouse profiling software, like all the manufacturers have today, in an independent profile producer. Huelight is one such individual today that sells profiles, but that's the only one I know of.

I think the profiling standstill is a little bit for the same reason of the DNG standstill -- photographers in general are not really interested. They're pleased with the current situation, and don't really have the knowledge of what's possible. I think most laymen think camera color sits 90% in the hardware and 10% in the profile, rather than the other way around which is closer to the truth. Camera makers have everything to gain from maintaining that myth too, especially those selling more exclusive cameras.

I think it's silly that cameras are made to have significantly different looks, and that you can't have a reproduction profile out of the box. How strange is it to expect that a camera can produce reasonably correct colors? The hardware can, but the profiling tradition doesn't allow it. Subjective looks is a good and convenient thing to have, but it can and should be added on top.

DCP actually has a wonderful layered approach which I use in full in DCamProf, first a matrix to get as close as possible to scene referred colorimetry, and then a LUT on top to refine that (HueSatMap tag). And then an additional LUT to add a look (LookTable tag). With RawTherepee the individual DCP components can be toggled on and off so you can choose if you want to use the colorimetric part only, with or without non-linear LUT corrections, and then toggle the look. What I've seen from Adobe's profiles they don't seem to design profiles this way themselves though, I don't know why. In ACR you can't toggle the individual parts anyway, so I guess it doesn't really matter from a user perspective, but seems strange to design the profile architecture this way and then not employ it.

An even better architecture I think would be to have just a colorimetric profile, which indeed would make all camera models look very similar, and then have look profiles entirely separate, which you can apply to any camera. The profile-does-it-all is legacy from the days when digital photography had to be packaged as an easy and familiar alternative to film photography, and choosing profile is like choosing film roll. I think we are past that period now though, and raw conversion could and should concentrate on being digital rather than mimicking an analog workflow.
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1996
Re: Yet some DNG comments (from a raw software developer)
« Reply #55 on: August 27, 2015, 04:53:40 PM »

but seems strange to design the profile architecture this way and then not employ it.

Adobe needs to design the UI to make the product usable by a lot of people, so between addition of some more (and more and more...) sliders/buttons/checkboxes and keeping it as it is ACR/LR actually is right, they gain a little market by adding those, but make UI more complex for hoi polloi... users aggravations do not worth that... RT tends to be overloaded by (useful indeed) features.

Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1996
Re: Yet some DNG comments (from a raw software developer)
« Reply #56 on: August 27, 2015, 04:56:48 PM »

An even better architecture I think would be to have just
just a SSF (or multiple SSFs) and make a tab (in raw converter UI) with controls allowing you to build a profile on the fly (and save as preset)... but again - that shall be hidden from a regular Joe in order not to cause the tears.
Logged

BrianVS

  • Full Member
  • ***
  • Offline Offline
  • Posts: 156
Re: Yet some DNG comments (from a raw software developer)
« Reply #57 on: August 27, 2015, 05:29:11 PM »

The problem becomes first to measure the spectral response for each sensor as implemented in the camera (accounting for IR cut filters, etc) , then to patch it into the DNG/TIFF-EP file. It might be easier to start an online database of spectral response curves for various cameras with measurements provided by end-users. Nikon, Canon, Sony, Olympus... do not provide the data so it must be measured. Anyone coming across files from the Leica M Digital cameras will have an easy time making up the curves as they are published. Same with my DCS200ir- the response curve is published. For most cameras: the curves are not published, and the cameras will be long inoperative when imagery from them is being recovered.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #58 on: August 28, 2015, 03:25:59 AM »

just a SSF (or multiple SSFs) and make a tab (in raw converter UI) with controls allowing you to build a profile on the fly (and save as preset)... but again - that shall be hidden from a regular Joe in order not to cause the tears.

Yes, indeed. I think preset looks are good because regular Joe won't be able to design them even if they had the interest (which they 99% of the users won't have). I've worked with subjective profile design a little lately, and it's hard. I think I can do better subjective profiles than both Phase One and Hasselblad which has been my "role models" when experimenting (mostly Hasselblad), but I've not settled yet and I'd say it takes weeks of field testing before one has a look settled. And while I don't think you need super-human color vision, one need to have healthy color vision and quite some training to be able to comfortably move in the space of subtle subjective color adjustments.

Hasseblad's Phocus actually has a design where the look is separate from the profile. Their profile is not colorimetric but I'd say it's "95% realistic", that is they don't try to make any particular strong look (like most others do), and I've heard that all their cameras look almost exactly the same despite different sensor hardware (like it should be), and then they have selectable looks on top like "nature", "portrait" etc which are simply presets of the raw converter adjustments tools. It's a bit too simplistic though, I think the subjective looks require different more subtle tools than the typical adjustment tools, so it would be better with a separate panel.

Lightroom is one of the more "consumer-oriented" softwares, where ease of use is the top priority, and I think that is the best commercial strategy. 99% of the users don't care that for example Capture One has better color adjustment tools (which indeed are much more difficult to use).

With proper design it would be possible to make a modern color rendering pipeline and still maintain the ease of use though. There's just this die-hard idea that digital cameras are just the same as analog cameras, just faster to get to the print.

As a profile designer it's much you can do to make a technical progress, with DCamProf I just have to live with the rendering pipelines that are out there, which means that I must provide features to build subjective look into the profile itself although I think that is not how camera color should work. Supporting Capture One has been the worst, as their pipeline is a real old-school mess. For example they split the curve between a user-selectable curve and a fixed curve in the ICC profile (to minimize color shifts the RGB curve causes), and there's a bunch of workarounds to get decent precision out of the integer-only v2 ICC profiles. With their hand-tuned profiles they get fantastic color though, and of course as long as you hand-tune the final result the underlying pipeline can be anything. The end user only sees the final result of course, and does not care how it's implemented.

("My" raw converter, RawTherapee, is a disaster in terms of user friendliness of course, I call it an experimental box for technically interested photographers. It has lots of legacy too and many design issues, so I don't hold it up as a role model in any way... it's a free open source software with programmers that come and go and spend some of their limited spare time on, so it is what it is.)
« Last Edit: August 28, 2015, 03:34:54 AM by torger »
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3231
Re: Yet some DNG comments (from a raw software developer)
« Reply #59 on: August 28, 2015, 03:43:12 AM »

The problem becomes first to measure the spectral response for each sensor as implemented in the camera (accounting for IR cut filters, etc) , then to patch it into the DNG/TIFF-EP file. It might be easier to start an online database of spectral response curves for various cameras with measurements provided by end-users. Nikon, Canon, Sony, Olympus... do not provide the data so it must be measured. Anyone coming across files from the Leica M Digital cameras will have an easy time making up the curves as they are published. Same with my DCS200ir- the response curve is published. For most cameras: the curves are not published, and the cameras will be long inoperative when imagery from them is being recovered.

Yes it's a political challenge. Someone's gotta measure it, and they won't get paid for it. I think Adobe is in the best situation for doing this. Maybe they already do it (to render their profiles), and they need to support new cameras anyway as their commercial success builds on that. They get paid indirectly as they sell ACR subscriptions.

The reason to embed it in the DNG is simply to make the format truly self-contained and archival, I think it's a natural part of a "digital negative". It doesn't necessarily need to be mandatory, but I think it should be a clear and documented option for the format. If DNG ever evolves into an archival standard I think it would be a huge mistake to not have this.

The idea is that the DNG should contain all information needed to open it in a future raw converter that may use an entirely different color model than those used in raw converters today, which as discussed already feels old (at least I think so). I think it's unwise to consider DCP to be a "future-proof" format for color rendition. The SSF is the ideal future proof information.

Today we have a few archival formats for the raw converter output, TIFF, JPEG, JPEG2000, and DNG has potential to be it for the input. Although I do see it as a wet dream, I don't think it's realistic to get an archival format for parametric raw conversion though, but of those three I think it's the least important too. Output first (we already have that), and then the digital negative. When those two are established and raw conversion has matured more than it has today, there may be a possibility to actually strive for an archival format of parametric raw conversion. I'd use it.
« Last Edit: August 28, 2015, 03:50:26 AM by torger »
Logged
Pages: 1 2 [3] 4   Go Up