Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: Fine_Art on June 01, 2014, 07:11:17 pm
-
I am a HDR newb. This was 3 .7ev bracketed shots out of a big pano.
I did the raw conversion into 32bit floating point in images plus. Screengrab to MS paint.
Snow tones seem ok? Darks Ok? Nothing looks nasty grunge to my eye. Opinions?
-
Here are the 3 frames in linear 32 bit FP for interest in discussion. Screenshot cropped.
-
It does not look like a HDR because the end result doesn't appear to be any different than a single capture?
-
It might not look vastly different from the brightest +.7 in linear, it is still plus about 1ev dynamic range over the D600 regular. The brightest is blown out in the snow even before gamma. I am using the darks from the brightest shot with a small gamma curve, not the typical 2.2. The snow is from the darkest bracket with gamma near 2.2 taking it close to 65535 in 16 bit.
-
When you shot the scene did you think it was a high contrast scene that justified the need for HDR? I have the D600 and in my limited knowledge of using Rawdigger then the D600 has about two and two thirds of headroom exposure before over exposure becomes a problem. Your image looks a little flat.
-
When you shot the scene did you think it was a high contrast scene that justified the need for HDR? I have the D600 and in my limited knowledge of using Rawdigger then the D600 has about two and two thirds of headroom exposure before over exposure becomes a problem. Your image looks a little flat.
I have no idea what you mean. To me blown is blown. If something is colored it might blow on only one channel so software can "reconstruct" to some level of acceptability. Sunlit snow is going to blow out fast. Yes, I saw it as a HDR scene for the mixed clouds with sun. I wanted a shot with clear control of the sunlit snow, I wanted a shot with detail in the shaded trees.
How the software decides what to map where, I have no idea. I know I fed it 3 images .7 ev apart. It gave me one output image to work with. I decided to output as 32bit FP .fit so I could manipulate it with lots of tones.
Maybe the trees look flat, I am undecided on the proper lightness for them. They need to be lighter without turning the rest of the image flat. I am trying to do that without using masks on the file. That needs more work.
-
I generally agree that this is not a good HDR candidate scene based on brightness. But OTOH, sometimes HDR on realtively low brightness scenes can give you a richer looking image and lower noise. Sometimes.
I notice that all your exposures are much too dark for a good HDR set. Your "middle" exposure is more like what one would want for the darkest exposure, and there is not an appropriate "bright" exposure that includes details in the darkest areas of the trees. Basically, your HDR set is overall about 2 stops underexposed.
-
Doing bracketing at ⅔ f-stop intervals does not make much sense for HDR, as almost any modern camera, and certainly D600, can easily cover such a dynamic range in a single shot.
-
Doing bracketing at ⅔ f-stop intervals does not make much sense for HDR, as almost any modern camera, and certainly D600, can easily cover such a dynamic range in a single shot.
You are right. I went back to have a look at the files in Nikon's ViewNX. The darkest file does fit everything in with a ETTR.
-
I generally agree that this is not a good HDR candidate scene based on brightness. But OTOH, sometimes HDR on realtively low brightness scenes can give you a richer looking image and lower noise. Sometimes.
I notice that all your exposures are much too dark for a good HDR set. Your "middle" exposure is more like what one would want for the darkest exposure, and there is not an appropriate "bright" exposure that includes details in the darkest areas of the trees. Basically, your HDR set is overall about 2 stops underexposed.
I may have confused the issue showing the linear data. This is the brightest file untouched in Nikon's ViewNX It is blown by about 1.3 stops. The darkest file is ETTR.
-
This is the darkest file. Sorry for the confusion, people are not used to seeing linear, that is my posting error.
-
I have no idea what you mean. To me blown is blown. If something is colored it might blow on only one channel so software can "reconstruct" to some level of acceptability.
Download RawDigger and you can judge more accurately if something is blown. You can't rely on the histogram on your camera or in LR or PS or a similar program.
http://www.rawdigger.com/
-
Shadow detail in the trees - or is it a lake reflection below the mountain? - is much too dark for an HDR, imho. A good HDR should show more shadow and/or highlight detail than presented.
Only 0.7 stop is too little. Bump that up to maybe +/-1 or even two stops for HDR as your Nikon has ADL (if you use their software to read it) to cover the smaller stuff. If something is too dark or two light, 2 stops might pull that back into a workable range at the expense of burying the other end in the wall of the histogram. The HDR software will fix that matter once it combines them all.
SG
-
2nd try. The trees look better. It's starting to skirt the edges of becoming a caricature of the original scene. My own tendency is to tone the contrast back a bit from here.
-
Your blackest black in foreground brush detail is around 20RGB in the green channel. Is that shot taken at night? Morning?
Try a notch curve in the area of 20RGB luminance to add definition.
Or find a slider adjust in your software that linearly shifts the entire image to the left viewed in the histogram (gamma encoded readouts NOT LINEAR) to establish absolute black and then apply a huge shepherd's hook shaped curve which most Raw converters already apply so you don't have to deal with editing such dark images.
You're having to keep going back and editing the image and still having it look dark at least as it shows on my calibrated display indicates you're not using the entire dynamic range captured and/or your software isn't helping you very well. I've taken far more dark, high contrast images than yours and normalized so it isn't that dark. Whatever software you're using is making it harder for you to achieve this.
-
One of my reasons for trying the HDR technique was the 'zero noise' process created here by our Spanish member, Guillermo. Lower noise, supersampled resolution, more tones in the data, everything seems to point to using multi-shot techniques. If I look at the tree line in the image that fits everything, the trees are limited in definition relative to the brightest shot. The brightest shot has the shape of branches. The darkest is a soft outline. I'm sure I was using MLU with a cable release so it was either a wind gust or the raw processor does a far better job with higher value data. I have to decide if it is worth the process of HDRing all the shots, getting them to a certain look, the stitching them. Its about an 8x3 pano so it's a lot of mucking around in software.
Is it worth getting into the habit of always using a multishot process for static subjects?
Image added.
-
Its about an 8x3 pano so it's a lot of mucking around in software.
Is it worth getting into the habit of always using a multishot process for static subjects?
Image added.
I don't have the patience nor do I ever find a need to shoot scenes that require stacked exposures. Your scene of the mountains can be done with one exposure.
Guillermo's HDR process is for real HDR scenes that attempt to retain detail in both low lit (about 2000 lumens) architectural interiors with a lot of smooth surfaces in the shadows of furniture that perceptually amplify noise while at the same time retain outdoor day lit (about 10K lumens) scenes in open windows.
Your landscape is lit by a giant ball of fire in the sky which provides quite a bit of full spectrum photons even in shadows for your sensor where very little noise is going to be present if not hidden within the texture of trees and grassy fields. You have a more advanced camera especially for low noise than my 2006 Pentax K100D DSLR, so I don't see why you have to go through so much trouble with a landscape scene that's not technically HDR.
-
With the much improved dynamic range of modern digital sensors (14.3 EV in the case of my D800 and D800E) I find that the need/benefit of HDR only emerges when the DR of the scene is such that I need to take 7 or 9 exposures at 1 EV intervals to get a good final image. (Actually, using Nik's HDR Efex Pro, I only use the odd-numbered images at 2EV intervals, but the auto-bracketing on my particular cameras only allow up to 1EV intervals !! )
-
I don't have the patience nor do I ever find a need to shoot scenes that require stacked exposures. Your scene of the mountains can be done with one exposure.
Guillermo's HDR process is for real HDR scenes that attempt to retain detail in both low lit (about 2000 lumens) architectural interiors with a lot of smooth surfaces in the shadows of furniture that perceptually amplify noise while at the same time retain outdoor day lit (about 10K lumens) scenes in open windows.
Your landscape is lit by a giant ball of fire in the sky which provides quite a bit of full spectrum photons even in shadows for your sensor where very little noise is going to be present if not hidden within the texture of trees and grassy fields. You have a more advanced camera especially for low noise than my 2006 Pentax K100D DSLR, so I don't see why you have to go through so much trouble with a landscape scene that's not technically HDR.
Your post makes sense in theory. I still have to deal with the experimental evidence of more detail in the shadows of the overexposed shot. Yes, one shot will do. More data does seem to capture it better. I also have quite a bit of experience using the multishot ISO NR built in to the Sony camera. It takes several fast shots (no mirror flapping) putting them together into a single jpg output. in many outdoor scenes it was the only way to get all the data in to one shot.
-
Sometimes, all you need is a few extra stops in one end of the exposure to create the image you would like, with the room to make the edits you like. The two basic routes are exposure fusion and HDR merge, with programs like ZeroNoise being a hybrid of the two. More recently, this hybrid method has been implemented such that one may output a "noise free" DNG, making this data amenable to a raw workflow (see LumariverHDR, for example). I know that Guillermo's approach included plans to write the output from ZeroNoise to DNG - LumariverHDR picks up where Guilllermo's approach left off. In cases where you only need a few extra stops of good data, you can get away with 2 or 3 images at +/- 2EV for this hybrid method.
Another approach that is really interesting but requires a little more of a convoluted workflow and understanding of the limits of the output (and a Canon camera) is the Magic Lantern DualISO module for supported Canon cameras. The supported Canon sensors have a design that the ML folks have exploited to read, simultaneously to a single raw file, alternating stripes (every 2 sensor lines) of the sensor at two different ISOs. This permits one to expose for one end of the scene and then use the DualISO module to acquire the other end of the scene simultaneously. For example, you can expose for the highlights in a scene (even ETTR exposure) at ISO 200 and set up the DualISO module to also acquire half of the sensor at ISO 1600. This results in better exposure of the shadows, with less noise. The single resulting raw file is then processed with a small application (cr2hdr) that combines the striped data into a single DNG. The method has its downsides (resolution, moiré can result) but is a very effective way to expand the useable dynamic range of your sensor AND capture moving subjects that, typically, produce alignment and ghosting issues in traditional HDR techniques.
kirk
-
Your post makes sense in theory. I still have to deal with the experimental evidence of more detail in the shadows of the overexposed shot. Yes, one shot will do. More data does seem to capture it better. I also have quite a bit of experience using the multishot ISO NR built in to the Sony camera. It takes several fast shots (no mirror flapping) putting them together into a single jpg output. in many outdoor scenes it was the only way to get all the data in to one shot.
So did this particular mountain landscape scene pass your HDR test with regard to seeing more detail in the shadows of the foreground trees and grassy fields?
I'm not seeing this in your posted final results, so I'm not sure what parameters you're defining as a passing HDR test.
You still haven't told us whether that is a daylight, morning, late afternoon or early evening scene so we know from the amount of light in the original scene if those shadows are suppose to be that dark and murky.
Why is your shot so dark and why don't you lift the shadow detail?
-
The shot was at 18:40 about a month ago. The sun now sets here at 21:30, it was probably setting at 19:30 a month ago.
You can see the 100% crop of the treeline near the bottom of the first page. If you look at the snow, it is sharper in the darker shot. The trees are more defined in the lighter shot (+1.3ev). Neither shot had vibration blur. I conclude the lack of fine detail in the trees of the darker shot (normal exposure) is from variability at the pixel data. The de-bayer has a hard time deciding what is what so it is more a continuous mash, to use the scientific term.
-
Here is a 100%crop from the other end of the same row, same pano. This time the in camera jpg just in case it was raw converter wierdness. Other people that are interested will of course test their own shots.
The +1.3ev shot has clearer tone definition in the dark trees. The 0ev shot has clearer snow (of course, a lot is not blown out). The sharpness of jpgs is less than raws. It's hard to say if there is a big difference in edge sharpness with the camera seeming to do USM..
-
Here is a shot showing good snow tones. +.67ev for ETTR. This is at 50% after deconvolution so it is quite sharp.
Then look at the same frame in another section that shows the dark trees in direct sun AND in cloud shadow in the same crop. The shadowed section turns to mush. Deconvolution really has nothing to work with.
-
Here is a shot showing good snow tones. +.67ev for ETTR. This is at 50% after deconvolution so it is quite sharp.
Then look at the same frame in another section that shows the dark trees in direct sun AND in cloud shadow in the same crop. The shadowed section turns to mush. Deconvolution really has nothing to work with.
It appears your deconvolution sharpening did work in the shadows. It's just that you can't edit what you can't see.
So I took the liberties of assigning a 1.0 gamma sRGB profile to a 16 bit version of the dark trees screenshot which opened up a ton of detail and applied Levels to add back definition with an additional 20a/20r USM in Photoshop. As you can see in the posted image below the trees and foreground brush are quite sharp.
If I can bring out that much detail in an 8 bit, gamma encoded screengrab with no stacked exposure blending, imagine what you can do on the original Raw version. And BTW the highlights didn't get brighter except when I applied USM which sent them to 250RGB.
-
It appears your deconvolution sharpening did work in the shadows. It's just that you can't edit what you can't see.
So I took the liberties of assigning a 1.0 gamma sRGB profile to a 16 bit version of the dark trees screenshot which opened up a ton of detail and applied Levels to add back definition with an additional 20a/20r USM in Photoshop. As you can see in the posted image below the trees and foreground brush are quite sharp.
If I can bring out that much detail in an 8 bit, gamma encoded screengrab with no stacked exposure blending, imagine what you can do on the original Raw version. And BTW the highlights didn't get brighter except when I applied USM which sent them to 250RGB.
How did you get a 16 bit image from the jpegs he posted ?
Because I am ignorant, I don't know how to do this:
'Assign a 1.0 gamma sRGB profile'. Could u link me to tut? Or even tell me what that means- is it a custom made camera profile? ICC profile?
By applying levels, I assume all you did was adjust white and black points and not mess with the gray point or exposure?
Thank you for helping those of us who are slow.
-
In PS, you can simply assign an existing profile to an image. If you have an sRGB profile with gamma of 1.0, you just Edit > Assign Profile... and choose sRGB Gamma 1.0.
These ICC profiles exist and you can add them to your computer in the appropriate location to make PS aware of them.
If you want to use them in the way described above, assign the gamma 1.0 version to an image and then convert to a standard (i.e., gamma 2.2 sRGB, Adobe RGB ,etc) profile to preserve the changes the assignment step made to the tone of the image.
You can modulate the effect by adding an adjustment layer (any tone adjustment layer, like curves, levels, etc.) to the layer stack, setting the blend mode to multiply and playing with the opacity prior to converting to the standard gamma profile.
Here is a free PS panel that permits you to assign a "False Profile" of various gamma - when you download the panel, it should come with various gamma versions of typical profiles that one might use in image editing.
http://bigano.com/index.php/en/freeware/100-software/288-cs-extension-false-profile-2.html
If you open a JPEG (8 bit) you simply change the mode to 16bit - then you do all of the assign profile jazz.
kirk
-
How did you get a 16 bit image from the jpegs he posted ?
Converted to 16 bit mode in Photoshop. It's what happens when opening a jpeg in ACR/LR set to 16 bit automatically. Just a precaution to reduce artifacts from applying such a huge brightening edit.
Because I am ignorant, I don't know how to do this:
'Assign a 1.0 gamma sRGB profile'. Could u link me to tut? Or even tell me what that means- is it a custom made camera profile? ICC profile?
Created it in Photoshop's "CustomRGB" under Color Settings. Why would you need to know how to do this anyway? Are you having trouble seeing shadows in landscapes in your Raw converter as the OP?
By applying levels, I assume all you did was adjust white and black points and not mess with the gray point or exposure?
After first converting to AdobeRGB, applied 50-60 on the left top black slider, 1.50 middle slider in levels. Eyeballed left slider for each RGB.
I really didn't see the point of explaining since the OP is working on a Raw file in a converter that seems to be giving him trouble allowing him to see and edit all that data in the shadows at the same time alleviate his concerns over noise.
Thank you for helping those of us who are slow.
Been doing it online for over 10 years for free. And you're welcome.
-
We are about to enter the third page of this thread and I am still confused as to what exactly is the issue we are debating? Are we debating anything? What was the problem, what are we agreeing or disagreeing about? Or we are just going in circles, stating the obvious and perhaps talking past each other?
-
We are about to enter the third page of this thread and I am still confused as to what exactly is the issue we are debating? Are we debating anything? What was the problem, what are we agreeing or disagreeing about? Or we are just going in circles, stating the obvious and perhaps talking past each other?
Something tells me you've already formulated your answers to those questions, Slobodan.
If you can't derive anything from this discussion, why do care? It's not all about you, ya' know.
-
Sometimes, all you need is a few extra stops in one end of the exposure to create the image you would like, with the room to make the edits you like. The two basic routes are exposure fusion and HDR merge, with programs like ZeroNoise being a hybrid of the two. More recently, this hybrid method has been implemented such that one may output a "noise free" DNG, making this data amenable to a raw workflow (see LumariverHDR, for example). I know that Guillermo's approach included plans to write the output from ZeroNoise to DNG - LumariverHDR picks up where Guilllermo's approach left off. In cases where you only need a few extra stops of good data, you can get away with 2 or 3 images at +/- 2EV for this hybrid method.
photoacute will output to DNG.
-
We are about to enter the third page of this thread and I am still confused as to what exactly is the issue we are debating? Are we debating anything? What was the problem, what are we agreeing or disagreeing about? Or we are just going in circles, stating the obvious and perhaps talking past each other?
Maybe HDR was not the best title for the thread. I called it HDR from the program's multiframe merge menu item. They are typically called HDR merge.
I used several exposures with some exposed to provide high tonal bits to the shadow portion of the scene. Does this improve on the dark data capture designed into the camera? The D600 has one of the highest ratings for DR and for color on DxO. Maybe this is asking for a lot! I think I will take the advise of increasing to a full 1 to 2 stop bracket. As long as I chimp to make sure I have a good ETTR frame, the rest is gravy.
The limit on being able to see the captured tones is likely the monitor. I use a HDTV, calibrated, with fairly high contrast. Are the gradations more noticeable? Maybe. I can easily see the darkest tones on a step wedge at DPR. Maybe we need higher bit depth displays.
I have a pano row 22000 pixels by 3000 that looks very good. :o I still have to work on the rest. Whether people want to waste the few seconds to capture extra frames is up to them. The data may not be valuable with today's output mediums, it may extend the usefulness of the shot into the future. Let me ask, when you look at shots you took with an early DSLR do you still see it as a good capture? Do you want to go back with your newer system to retake the scene?
The thread can end or we can discuss the value of large data capture.
-
...Let me ask, when you look at shots you took with an early DSLR do you still see it as a good capture?...
They are as good as it gets. Actually, they are, like good wine, even getting better with age. Not because captured photons alter their state with time, but because RAW processing has made huge advances, together with our ability to use them more effectively. I am often surprised how good my old Canon 20D files are in Lightroom 5.
-
I used several exposures with some exposed to provide high tonal bits to the shadow portion of the scene. Does this improve on the dark data capture designed into the camera?
Those shadow tones will suffer more by their distance, resolution and lens accutance than from theoretical tonal bits captured by stacking multi-bracketed exposures. What I saw editing your screengrab indicates to me you've captured more than the eyes could see at that scene. I can also say that of my 2006 Pentax 6MP DSLR capturing at lesser distances. But currently I'm satisfied with my prints.
Resolution aids far more at capturing all those tiny grass strands and pine tree leaves 100's of feet away in your pano landscape as long as the lens can sharply map it to the sensor. You may have enough pixels to uprez that pano to print to fill a 25ft. wide wall after you're through. The issue now is the viewer will have to stand way back about 25ft just to take it all in. So what's the point in capturing extra data that can't be seen even in a print? Don't you feel it's overkill with all this work on top of your camera's advanced resolution/DR capture capability?
Let me ask, when you look at shots you took with an early DSLR do you still see it as a good capture? Do you want to go back with your newer system to retake the scene?
Sure, I'ld like to be able to afford a newer and more advanced camera system, but if I had that system, I certainly wouldn't go to your lengths at producing a pano and instead let the camera do most of the work. But that's my POV.
-
The 85 on the 6016 pixel wide sensor gives you about 4x 20/20 vision. The 21494x2816 pano row is about 90 degrees wide at 4x magnification compared to being there with 20/20 vision. If you remember the old "Is it live or is it Memorex?" ads, Memorex or live, are both poor substitutes for this image. ;)