[Solved] hdr workflow - ask for guidance

gadolf · May 20, 2018, 8:27pm

Hi.

I’m struggling to reproduce @Carmelo_DrRaw in this post, which I quote:

I prepared a scene-referred starting point by merging the RAW files with HDRMerge and saving a neutral processing in linear Rec.2020

I saved the following exposure steps from the HDRMerge output: -1, 0, +1, +2, +3

from the scene-referred version I saved 5 images at different exposure values (from -1EV to +3V with respect to CRW_3012.DNG), which were then processed with enfuse. The enfuse output looks flat, but provides a compressed dynamic range with limited suppression of the local contrast.

I took other three images, then followed the exact outlined steps, and then opened the enfuse output in Darktable. The problem is that I can’t properly handle the highlights, that is, I’m not able to recover detail in the sky. There’s only a big white sky.

Regarding the shadows, everything is ok, and I get much better results in tone mapping than I would if I was working with a non-compressed image.

What am I doing wrong? Isn’t this workflow applicable to any image? Shouldn’t I be able to manipulate the compressed information in the highlights as I do in the shadows?

As a side note, I also tried working with different exposure values in the enfuse step, like -2, -1, 0, +1, +2, and -3, -2, -1, 0, +1, but then I couldn’t get the same results in the shadows and, although I could recover more detail in the highlights, I still couldn’t achieve the contrast I’m aiming at.

Follow the three original raw files:
CRW_3486.DNG (17.8 MB)
CRW_3488.DNG (17.8 MB)
CRW_3487.DNG (17.8 MB)

hdrmerge output:
CRW_3486-3488.dng (34.8 MB)

enfuse output:
https://filebin.net/7e3jvzvruwxkvlv0

CriticalConundrum · May 20, 2018, 11:51pm

If your sky is clipped in the raw then there’s nothing you can do recovery wise. Maybe get another picture of a nice sky and shop it in.

gadolf · May 21, 2018, 1:03am

@CriticalConundrum Which one?

CriticalConundrum · May 21, 2018, 1:50am

Apologies, I thought you said your RAW was clipped.

If my edit is anything close to what you want I may be able to help you out

afre · May 21, 2018, 2:49am

The issue is your use of enfuse and expectation of highlight recovery.

If you read the enfuse manual, you would see that there is much that you could do with it, including choosing which parts to include via masking and alpha.

If I recall correctly, your raw files were on the dark side. You can only recover what is there to begin with and not create something out of nothing. When you have multiple frames, you have more data. But more data where? In your case, the data density is still in the shadows. Hence, it makes no sense to export other exposure value sets from HDRMerge. In fact, you only export exposures where the various regions of your image look right.

Some photographers don’t approach it this way. They take creative license and synthesize or replace the sky altogether.

PS If you would like to gain more experience working with highlights, try tackling this [PlayRaw] 4: the Irish musician. Try to balance the light and retain detail in the lights. It took me a few tries and my highlight management skills improved as a result.

gadolf · May 21, 2018, 10:38pm

@afre My point is why this enfuse result (with default settings applied) reveal details on the sky,

… and this one doesn’t (also using enfuse with defaults)

given that both +3 EV frames show completely white sky

and

In other words, I’d like to check if I’m doing something wrong before considering the alternatives you’ve pointed out.

gadolf · May 21, 2018, 10:39pm

@CriticalConundrum Yes, it is.
Would you like to share your workflow?

afre · May 21, 2018, 10:55pm

I don’t know exactly what you mean but my guess is that the two images are very different in terms of content, contrast and proportion. Using only the defaults is like opening a raw processor and doing nothing. That won’t work for every photo.

If you are still following @Carmelo_DrRaw’s example, ask him to elaborate on how he arrived at his results.

Who knows, maybe this is a tutorial in the making!

gadolf · May 21, 2018, 11:44pm

Maybe that’s what’s happening here and I will end up splitting this workflow into two parts, one for the ground and one for the sky, as you mentioned before.

Meanwhile I’ll wait a bit more to see if some new idea comes into play.

Thanks anyway for your support.

CriticalConundrum · May 22, 2018, 8:35am

@gadolf It’s really fast. I export the HDRmerge raw with a neutral tone curve into a 32bit float format. I then open this in a photo editor and duplicate into 2 layers for shadow and highlights. I apply the S curve for highlights ignoring the effect on shadows and blend it into the highlights region using blend if. Then I do the same for shadows. Ideally you try to get the contrast about equal across the board and then you can do one last S curve for everything.

I did in fact finish off with a Tonal Contrast effect from the NIK Collection, which is free, but honestly it’s kinda like icing on the cake.

gadolf · May 22, 2018, 10:16am

@CriticalConundrum On which color space?

EDIT: I agree, working with the hdrmerge output is easy. I got something like this, without layers:

But with that I loose all compressed dynamic range provided by enfuse. In the workflow I’m trying to achieve, I create images with different exposures from the hdrmerge output and then merge them with enfuse. With that, you compress all dynamic range into a single image, which could then be edited, revealing tonalities at taste. Also, by doing those steps on a wider gamut color space, you make sure you don’t loose any data. See this post.
Working with enfuse gives me much better results on the shadows:

My point is also achieving the same results on the highlights, which I thought I would be able to do in the same edit session. But, as @afre said, I probably will need to split this workflow into three parts, one for shadows, one for highlights and then merge the two outputs.

Carmelo_DrRaw · May 22, 2018, 6:58pm

Sorry for being late to this party, but life kept me very busy recently…

This is my “best” result with enfuse, obtained in the following way:

I saved images at 1EV intervals, from -3EV to +3EV, in perceptually-encoded sRGB and 16 bit TIFFs
merged the images with the following command:

--exposure-weight=1 --saturation-weight=0 --blend-colorspace=IDENTITY --gray-projector=luminance --depth=r32 --output=CRW_3486-3488-enfuse.tif CRW_3486-3488??.tif

Here is the result:

The result is quite flat, particularly in the sky…

Recently I have started to play with another method, which is based on the bilateral filter. This method is not new, and there is quite a lot of scientific literature on the subject (see for example this and this articles). It boils down to performing a two-level decomposition of the image, where a blurred version of the image is subtracted to the original one to isolate the high-frequency details. The contrast of the blurred image is then flattened to compress the overall dynamic range, and finally the high-frequency level is added back to restore the local details.

The key is in the blur operator: when using a simple gaussian kernel, the result will show visible halos around strong edges. This problem can be mitigated, or even practically removed, by using an edge-aware blur operator like the bilateral filter. Such blur operators are able to “distinguish” between fine-grained texture (that gets blurred) from hard edges (that are left unchanged). As a result, the high-frequency level will contain the fine-grained details, but not the hard edges, which are kept in the blurred version.

Here is what I could obtain with a simple application of the above method:

The result gets even better after restoring a bit of contrast to the merged image:

Here is a PhotoFlow preset that automates the procedure:
SH-recovery-bilateral.pfp.zip (1.8 KB)

afre · May 22, 2018, 10:12pm

Looks better than your enfuse output. I came up with and experimented with a similar method but it didn’t perform well. Guess I will revisit it sometime.

gadolf · May 22, 2018, 11:58pm

@Carmelo_DrRaw Thank you very much for taking your time and sharing your “new” workflow!
I loaded this image, applied your preset, and exported to a tif file. I did the same to other two images that produce a panorama.
Unfortunately, Hugin doesn’t seem to like these files: after setting up the parameters (finding the points / geometric and photometric optimization), when I click on the GL button (to start building the panorama), it crashes. I tried this for rec2000 tifs as well as for srgb’s. Reinstalled it but nothing.
When I have time I try to test this Hugin/Ubuntu 18.04 on other panoramas to see if the problem is with this new Ubuntu.
Thanks once more!

afre · May 23, 2018, 12:52am

Try (1) deleting Hugin’s preference files and (2) reading its logs.

gaaned92 · May 23, 2018, 1:21pm

my try

HDRmerge
Rawtherapee with fattal and some tweaks

Not sure if HDR workflow is useful in this case.(only 1 EV between darker and brighter photo)

CRW_3486-3488.jpg.out.pp3 (11.4 KB)

yteaot · May 23, 2018, 7:21pm

I do not understand the whole debate. I would open the darkest photo (3488) to
raw-editor and adjust photo whit it. Where’s the beef?

gaaned92 · May 23, 2018, 7:28pm

agreed

Carmelo_DrRaw · May 23, 2018, 7:54pm

One can always edit the darkest photo alone. However, merging multiple exposures gives cleaner shadows with much reduced noise. While this might not be crucial in this specific case, it is at the basis of any HDR workflow (which is the initial topic of this thread).

Carmelo_DrRaw · May 23, 2018, 7:55pm

In which format did you export the images? I am not sure that all Hugin versions are compatible with 32 bit floating point tiffs…