For #1: the DNG Profile Editor doesn't currently let you "edit by the numbers." One thing we're considering adding in the future is the ability to specify your own target values for the chart, which would let you optimize for your specific chart. On the other hand, typically ColorCheckers are quite close (tight tolerance); I think any variation in lighting you may have when shooting the chart would actually lead to greater differences than ones obtained between charts.
For #2: yes, the Chart Wizard feature (used in the Chart tab with a ColorChecker) is meant to do auto-calibration in the same way that the previous CR calibration scripts did. The DNG Profile Editor optimizes all color patches simultaneously.
It's not currently possible to change the reference Lab values.
Thank you very much, Eric
>For #1: the DNG Profile Editor doesn't currently let you "edit by the numbers." One thing we're considering adding in the future is the ability to specify your own target values for the chart, which would let you optimize for your specific chart.
It would be nice for reproduction purpose. In this way there will be no problems for #2
>On the other hand, typically ColorCheckers are quite close (tight tolerance); I think any variation in lighting you may have when shooting the chart would actually lead to greater differences than ones obtained between charts.
Anyway, I think that this ACR release together with DNG profile editor are really groundbreaking tools for photographers and color geeks!
thank you for a great new tool! However there's still room for improvement ;)
First, I second the request to edit colours "by the number", just like in Photoshop's Color Picker tool. Pushing sliders or moving an arrow across the colour wheel is simple and intuitive but not very accurate. I guess you have planned that for one of the next releases anyway, haven't you? ;)
Second, I'd like to see a way to automatically create a profile that resembles the in-camera JPEG colour rendition. When shooting in RAW+JPEG format then one could point the Profile Editor to a pair of JPEG and DNG and then a DNG profile would get created that matches the DNG rendition to the JPEG's. Would such a feature be possible? I guess many would love to see it---even when hard-core colour theorists may roll their eyes. I often shoot RAW+JPEG and will use the JPEG files mostly and the raw files only for those shots that need major tweaking. In this use case, matching colour rendition would be really helpful.
I would also like to know what colour space the editor is working in. The RGB values do not match my ProPhoto colour space ones with my Gretag CC. Not even close.
The way that I alway got the best results in matching (in my case) the Nikon NX colours with the Tindemans calibration script was to actually edit the target RGB numbers right in the script. Aiming for values measured in NX for each patch in ProPhoto and then let the script do it's thing.
I have not figured out a way to do this kind of adjustment with the DNG editor yet. In order to do so we would need to know the working colour space and be able to edit the target values that the colour table is aiming for during the automated process. I would think this would produce similar results but since it is able to target a narrower swath of colours it should be even better.
I am not sure if I am making myself clear in what I am asking for. Ask any questions if clarification is needed.
Olaf, what you're asking for with auto JPEG matching is not entirely possible because of the auto processing typically done on in-camera JPEGs, especially when image-dependent features like D-Lighting, vignette correction, etc. are enabled. These significantly alter tonal balance and in some cases color. And since they can vary from image to image, this becomes a moving target.
Chris, the reason why you're likely not seeing matches numerically is because the RGB readouts in PE are scene referred (i.e., it's linear: no tone curve or gamma encoding has been applied). The primaries are ProPhoto RGB, but there is no 1.8 gamma encoding.
When you use the chart wizard feature in PE, it is actually optimizing each of the 18 color patches using internal reference values (spectral data for several charts, averaged together). So you should not need to enter target values 1 by 1, unless your chart is highly unusual.
Thank you for the explanation. That makes a lot of sense regarding the linear gamma.
As far as the chart wizard feature goes, that is exactly how I thought it worked. Same as the Calibration scripts. They target an internal known value for each patch. This give accurate/ neutral colour response.
What I am referring to is if I want to target something other than accurate values. With a script I just went in and edited the ProPhoto RGB target values for the two colour temps based on colour point samples from Nikon NX (or any other program you want to match). Then the script is doing it's best to match the particular parameter set I was using in NX/other editor.
That is what I would like to be able to do with the DNG Profiler as well. I'm not sure if the internal target values are stored as linear values as you described above or not. This may complicate the matter.
Is there another way to do what I am asking other than trying for a visual match which is slow and not particularly accurate?
As of now my modified script ACR calibration is doing a much better job. I would really like to see if I can improve on it.
Thanks for taking the time to answer questions.
Umm ... the more I play with the Profile Editor, the more confused I seem to become.
Imagine an landscape shot with a lot of blueish and yellowish greens. Maybe I feel ACR is rendering the blueish greens fine but the yellowish greens too yellow so I'd want to tweak them. I'd load the shot's DNG file in to the Profile Editor, pick a green I'm unhappy with, and drag it a bit towards a more blueish green. Now what exactly would happen?
If I understand correctly, only the particular colour that I actually picked will get shifted; colours that are close to it will remain unaffected. Wouldn't that distort colour rendition substantially? Do I need to pick several kinds of yellowish greens and shift all of them, in order to improve the rendition of natural grass and Spring foliage (as opposed to artificial monochrome patches in a test chart)? If so, how many?
When tweaking the greens in the Color Matrices tab then
greens would get shifted, right? The yellowish ones as well as the blueish ones. That would not be what I want. Is there a way to see, and possibly to control, the "effective width" of the colour range that gets affected by a single tweak in the Profile Editor's Color Tables tab?
And then there is another, related question: What is the size of the Profile Editor's eyedropper tool? One by one pixels? It seems hard to pick a colour in the presence of colour noise. Can I increase the tool tip's size to, say, 3×3 or 5×5?
In your example, add a sample for both the colors you like and the colors you don't like. Only adjust the sliders for the colors you don't like. The samples in the colors you like will "pin" the results at those points in the final interpolation.
Thomas, thank you for your quick answer!
Your suggestion to pin down some colours while adjusting others implies that my notion of "shifting precisely the selected colour only" was entirely wrong. And furthermore, there is no fixed "effective width" of a single tweak; instead, the width of the effect depends on the other tweaks. I just discovered the "Show Affected Colors" option and what it does, and that nicely settles the issue for me.
Correct, Olaf, the extent of a color adjustment is __implicitly__ determined by the set of control points you've specified. Tutorial 1 of the docs elaborates on Thomas's explanation of "pinning down" colors:
The 'Show Affected Colors' option does indeed give you a visualization of the extent of an adjustment. You'll see that the extents change as you add/remove adjustments. Hope this helps.
Chris, I now see what you're getting at. Sorry, this version of PE can't do that: as you say, you'd need to be able to specify custom target values. For now the only way to do it is visually, which isn't that bad if you can bring up the target image in the background and place the image window in PE next to it, so you can see them side by side. Then use the chart wizard to create the initial set of 18 color adjustments, just so you don't have to manually click on each one. Then you'd need to adjust the sliders for each one to get you closer visually to the target. (keep in mind that unless you have a wide-ish gamut display, the last patch in the 3rd row will likely be out of gamut on the display).
But for now no automated way.
Before starting, I want to say that I had cataract surgery last month. The vision improvement was almost instantaneous and dramatic. I can now state that color differences in individuals can be much greater than expected. It is hard to express how big the difference is in my own vision, let alone between two other individuals. Any color assessment tools must accommodate numeric as well as visual comparisons. That said, the healing process has also affected how much time I can spend comfortably in front of my screen. I hope my comments will be taken constructively.
As others have commented, I also feel that the CC24 Lab target values being used should be documented. Published values from Gretag, Lindbloom, Babbage, and others disagree. Without the target numbers it is difficult to determine accurately the calibration result.
My first attempt seemed to go smoothly, but when I verified the results with ReadColors.jsx (my script) the numbers degraded slightly. I was using the Gretag target values, so these may not be what Adobe is using. Also, I had already calibrated ACR for my D3. The doc seemed to imply that the ACR tabs would all default to zero and that was what displayed in the Color Matrices panel. But that was not what showed in the ACR panel. So I tried again, setting all ACR sliders to zero. This calibration attempt was a disaster, numerically and visually. Now I am confused. What is the relationship between these and what are the recommended user actions? In each case, I exported the DNG profile and selected it subsequently in ACR to re-open the image.
Next, I played with the Color Tables panel. I set a watchpoint on the blue patch. As long as I hold the eyedropper over the blue patch the Lab (and other numbers) track. But as soon as I move the mouse to the sliders, the numbers disappear. If I move the hue and saturation sliders the image and the patch sample both change. And, yes a little arrow shows in the color wheel. But after moving the eyedropper back, none of the numbers have changed. The lightness slider is even more bizarre. Obviously no arrow (z-axis) but the image brightens and darkens (OK), without any corresponding change in the patch sample (watchpoint) on the right (not OK). And again, no change in the numbers at all. It would be very helpful if the target values were shown here and if the currently selected watchpoint that would track the numeric changes. It would also be nice if the watchpoint could show the target color as well as the initial and changed image patch colors.
I dont understand the purpose of the Tone Curves. I understand gamma and Adobe linear, but these do not correlate to the ACR tone curves. And again the image changes but the numbers do not. Since we cannot set watchpoints in the neutral patches it is very difficult to assess any changes the user might make.
Next, I used the Chart panel to Create Color Table. When I return to the Color Tables panel, there are the 18 color (no neutral) watchpoints. I can see that blue has changed in the watch point and the arrow in the color wheel. But the numbers are still the same as before. Scrolling through the watch points I can also see that the hue and saturation sliders have changed. But the lightness slider did not change in any of the color patches. I anticipated that this would be the major improvement over the ACR sliders. Is this just not in the calibration algorithms yet?
This is a good start, but until I know what the target values are and what I should be doing about the existing ACR slider settings, I am at a stopping point.
Cheers, Rags :-)
Read the PE tutorial if you haven't already. It's pretty straight forward.
What I gathered from it is that the PE profiles generated act like assigning a profile in PS only not the ICC type. The preview changes according to the adjustments applied but the linear source data stays the same. Correct me if I'm wrong on this.
PE is just a preview generator that creates INPUT profiles so the user doesn't have to work directly from the preview generated by the dark and flat linear data off the sensor though it does give this option.
When viewing the default preview generated by the default settings profile in ACR your RGB data readouts when editing are OUTPUT=ProPhotoRGB or whatever working space used.
If wanting to go by the numbers matching two targets in PE where one of the targets was the visual source that is known to be accurate, I wonder if using something like Apple's Digital Color Meter would work. I don't know if Windows has this yet.
The user would make adjustments in PE matching the DCM readouts taken from the accurate source target. These numbers would not be accurate to a measured reference data file because they're generated off the video card LUT but at least you've got a visual preview and those numbers to go by.
The RGB/Lab/HSL readouts in PE don't currently allow you really to 'edit' by the numbers. They're only there for the purposes of selecting a color to edit (i.e., color control point selection). At the moment the main profile editing technique available is visual, with the exception of the automated chart wizard feature. We didn't include full edit-by-the-numbers capability for this first round because my initial tests found it confusing: these are literally input profiles in that they operate on the linear scene-referred data, whereas all the numbers you're used to thinking about (including the numbers for a reference CC chart) are all output-referred. So there's a disconnect there that will take a bit more work to resolve nicely from an editing interface point of view.
The DNG 1.2 profiles (including ones created via PE) are intended to be used with CR defaults, in particular all Calibration sliders set to zero, Brightness 50, Contrast 25, and Tone Curve = Medium Contrast (and no Parametric Tone Curve). With these tone curve settings you'll be getting exactly the tone curve specified in the profile. If you zero everything out in CR (as you tried doing) then you're going to see something flat and dark because you are effectively applying a darkening tone curve on top of the profile.
The 2nd tab (tone curve) in PE effectively controls your default tone curve. It's a way of folding in your desired tonality into the profile, so that flipping between profiles becomes a handy way of setting a desired starting point for color but also tone.
My question is why so much layering of tone adjustments in rendering Raw data?
First you have the ACR default contrast, brightness and shadow slider settings then the medium contrast tone curve among the other curve styles provided. And the linear curve setting in ACR doesn't give the same response as the one in PE. What part of the ACR's default settings is part of PE's or are they combined on top of each other?
Then underneath this you have that huge arching curve in the DNG profile that corrects the REALLY dark linear sensor data and then on top of that the same regular contrast curves as in ACR. This linear correction curve is understandable because I tried to edit that in another Raw converter that gives that option and it was impossible to tweak especially shadow detail.
Is all this layering of tone adjustments to prevent artifacts from kicking up trying to make the dark sensor data of a scene look more like human visual response?
Also is the DNG PE profile used with DNG converted images the same one used with non-DNG Raw images in ACR?
Oh, I forgot to add.
This is one of the coolest tools you all have provided. Great job on this.
You've got my gearhead a churnin'.
Ultimately there is only 1 tone curve being applied.
But the tone curve is so important to controlling the final look of the image, there are lots of ways of editing it. You can keep things simple and just use Brightness/Contrast. But if you want more control use parametric and/or point curves. But like I said ultimately there's just one curve.
Not sure I understand your question about the DNG PE profile used ... a raw file has a camera profile, which you can select via the Calibration tab in CR/LR. Let's call this camera profile X. When you turn it into a DNG, then open this DNG in PE, PE will set its base profile to X. That's it.
Eric - Thanks again for responding. Custom target values is exactly what I am asking for. And to have them in a stated color space (ie ProPhoto). I'm not sure how much more difficult this is because of the linear gamma.
This will allow users to make their own versions of the manufactures modes or settings. For example, let's say I like the Nikon look with their low contrast curve, normal saturation and Mode II. All I do is open my GTMB CC shot in Nikon NX2 with those settings in ProPhoto. Note the RGB values of each patch. Then target them with the DNG profile editor instead of the default values. Presto we have a new default in a second.
Doing this visually (which I use to do before I figured out I could just modify the script) is slow and does not provide the same results.
I'd also like to say that I think this is one of the most interesting things I've seen come along in a long time. I'm really hoping this develops into a fully realized tool. I love using ACR and Bridge for my workflow and this would pretty much assure I could acheive any default look I would ever want.
This camera profile X from any raw file in question before conversion to DNG is embedded by the camera manufacturer? or is it from ACR's default settings?
About the huge arching "RED" colored curve as seen in the PE tutorial, where is this from? PE? the manufacturer? or ACR defaults? Is this huge curve being applied underneath the regular curves and slider settings?
I ask this because in ACR you can't turn the big curve off by selecting "Linear" in the curve dialog box. The preview doesn't go that dark as you'ld expect from turning off such a large curve at least not from my normally exposed raw files.
Chris, thanks for the feedback. We'll keep this in mind. I do understand what you're saying.
Tim, the profile embedded in the DNG and associated with the image could come from many places. For example if you're shooting with a Pentax or Ricoh camera then the camera will embed a profile in the DNG. If you're taking a non-DNG raw file and converting it to a DNG, the embedded profile will be the default Adobe profile for your camera (currently ACR 4.4) unless you've manually set it to something else in Camera Raw or Lightroom.
That red curve is the base tone curve, which comes from the base profile (specified in the Color Tables tab). You will get exactly this curve in CR/LR if you leave your CR/LR tone settings at the defaults (i.e., Brightness 50, Contrast 25, Point Curve = Medium Contrast).
BTW, the reason you don't see the preview go dark when setting point curves to 'linear' is because you probably still have brightness set to 50 (or something like that). Try setting it to zero and you should then see the image go pretty dark.
I really love this feature and I'm planning on messing with it a lot.
I do understand the requests for the ability to input custom RGB targets, but for me the auto-calibration feature using the ColorChecker will probably meet many/most people's needs... assuming they own a Gretag Macbeth 24 patch card.
Unfortunately I own a Perfect Pixs Natural Color 30 patch card. Any plans to extend the ColorChecker cards supported?
If not, then I also would like the ability to input custom RGB targets and then I can manually use the 30 patch card to point to specific custom targets without trying to visually make the changes to every color patch. (And on that note, in principal I prefer the 30 patch card because it specifically has numerous "skin-colored" patches).
Love the work,
I've noted your request, but at this time only the CC 24 is supported. (The Chart Wizard feature was mostly designed for folks familiar with the existing CR calibration scripts, used with previous versions of CR.)
While I like the new feature more, the difference is the CR calibration scripts could be edited and were available for a number of different calibration charts. The calibration scripts were actually the reason I bought the Perfect Pixs chart.
In any case, I love the tool, I just wanted to put my vote up there too :)
So after almost a year is there any word on adding the ability to target user specified Prophoto RGB values in a new version of the Profile editor?
I'm afraid not.
Any explanation? Is there a technical reason why this is difficult?
Would you say it will never happen?
I've spent some time trying to access precise of profiles created with PE.
As I know, PE doesn't change tonality of an image, but only hue and saturation. So, to make accession, I created a profile with ColorChecker chart. Then I opened the same image with ACR and set blacks, brightness, contrast to zero, increase exposure and correct tone curve to get the right values for grey patches. After that I measured RBG values for red, green, blue patches and compared them to original values of chart. As a result, I saw that DeltaE* is up to 6 units!
The primary reason is that you are effectively measuring color differences at the end of the process (i.e., output referred color) whereas the DNG PE's Chart Wizard is attempting to do reasonably accurate scene-referred color matching.
PIXIllate, it is not a question of difficulty, but rather of priority. This is not to say that your request is unimportant, but that there are other areas of image processing that require stronger attention at the moment.
Sorry, but I didn't understand your message. What is the right way to compare measured values with referred values?
Unfortunately there is not a straightforward way to do this comparison numerically, at least not meaningfully. You can certainly measure the patch values as output by Camera Raw and then compare them to reference values, but as you've noticed, they may be off. If you are getting visually unpleasant results, you can try editing the color control points as noted in the DNG Profile Editor tutorials.
The purpose of DNG Profile Editor is to provide accurate colours, not pleasant results. Or did I miss something?
Do a color number experiment in Photoshop. Create a new AdobeRGB document and convert to the default US Prepress Web coated SWOP v2 CMYK profile. Create a color fill by entering 100% cyan in the Color Picker. Once done convert back to AdobeRGB using the default Relative Intent. You'll note that the red channel clips to 0 when doing this but the preview of the color doesn't change.
Now go back into the Color Picker and sample this cyan while in AdobeRGB and raise the red channel from zero to as far as you can go until you see a change in the appearance of the cyan sample. On my system I have to go as high as 78 levels before I see a change. Does it still qualify as a SWOP certified cyan? It sure looks like cyan.
The point I'm making here is if there is this much wiggle room in defining what a color looks like according to the numbers (and this particular cyan was created from Lab measurements with a spectrophotometer in a controlled lighting environment) then how can you rely on any number system to nail the color exactly in number and appearance on a consistent basis from a device as dodgy and uncontrollable as a digital camera?
In short you'll do far better and faster if you let your eyes do the rest of the color judging. The numbers are only going to get you so close because a digital camera isn't a copier nor does it operate like our eyes or a spectrophotometer which defines the actual numbers. One small shift in exposure and contrast at time of capturing a scene that's off from the base conditions established at the time of creating the PE profile are going to make the numbers go as far off as the red channel in the cyan experiment with no change in the preview. Delta E numbers be damned because you won't see the encoding error when it happens. You then have to decide what's more important, having a great looking image or exact numbers.
Use your eyes. It's a lot quicker than relying on software to do all the work.