Testing your HDR applications

Hi folks,

I like to see how well various image processing applications can cope with merging and tone mapping high dynamic range scenes. I prefer to get the cleanest merge possible with data that encompasses the full dynamic range without too much noise in the shadows and full detail in the highlights, as well as good color rendition throughout the range. Important aspects of the performance of an application also include deghosting and alignment, although I tend to shoot on a tripod.

That said, I have found a couple of test image sets that may be useful to anyone interested in finding out how well their applications can cope with the typical kinds of problems and situations that will inevitably occur when shooting and processing HDR scenes. At the end of this post I will include a link to a Dropbox archive (ZIP) with the raw image files (one from a Canon 5DII and one from a 5DIII) that you can play with to see what the limitations might be in the applications you use. That is not to say that those applications are "bad" but I have found that these two images sets challenge a lot of different aspects of HDR imaging and can be useful in exposing those limitations. Some of the "limitations" are, I will admit, minor in the grand scheme of things but nonetheless may distinguish okay from good from really good.

Attached are the finished scenes. It is not important in this post to "reveal" how I came up with the finished scene in each image, although I used a very straightforward processing workflow.

Raw images > merge > save "master" file > tone map > finish in PS.

The ways that you can do this workflow vary over the range of applications and personal preferences. I put each image set through several popular HDR applications and found that most of them suffered, to some degree, from various issues in both scenes, with the exception of the application that I used to generate the posted images. To avoid the impression that I am being cryptic, I found LumaRiver HDR to be the reference application for merging and initial tone mapping of most HDR scenes. Unfortunately it is Mac-only. There are promising PC-only applications that will probably work equally as well, but I do not have a virtual machine on my Mac capable of running applications like SNS or Oloneo (is that still being developed?). Hopefully, this thread will provide feedback from forum members who use the test image sets and run them through whatever applications they use, and critique each based on a standard set of test images.

Okay, enough BS. The attached images contain two scenes. In each scene I provided three thumbnails to give you an idea of the exposure range for the source raw images.

Scene 1 is a very contrasty indoor scene with exposure to a brightly lit outside sunny day through a window. This is a common scenario for people shooting interiors. In this test scene, I did not provide any artificial lighting, other than the ambient glow of the computer display (which is calibrated to 90 cd/m2). Normally, I work with the blinds closed - in late afternoon the sun from the west falls on this face of the building I am in, producing brilliant light when it is sunny out. The window itself has some sort of plastic film (maybe some old UV film) which is partially falling off. The desk falls in the contrasty light and shadow of the sunlight and under the desk is dark, with little ambient light bouncing into the far reaches under the desk. It is amazing that our brains can see all of this stuff and perceive this kind of dynamic range. The trick here is to be able to extract detail in the area under the desk while maintaining midtone and highlight detail, and getting color rendered correctly (or at least pleasingly) across the tonal range of the image. There are color checker references in the image if you want to gauge your tests on them. Also, the very sharp edges of the very bright portal of the window next to the very flat areas of the walls expose issues with haloing in the tone mapper. The flat walls also help expose issues with rendering gradually changing areas of tone where a masking type algorithm might get a little inaccurate.

Scene 2 is a backlit landscape scene, typical of what you might shoot when outside. I chose to position the camera facing the unlit side of the tree and shoot into the blue sky to capture this mildly high-contrast scene. It was windy when I was shooting - I was able to use high shutter speeds so the tree leaves and branches did not blur, but ghosting is an issue, especially in the complex oak leaves overlapping each other and the evergreens in the background. The gentle gradient of the blue sky, and how it is rendered throughout the scene is a really good test of how the merge algorithm of an application handles the transition at clipping across the range of images. The lens I was using (Zeiss 50mm macro-planar) starts to flare internally and contrast is lost in the lighter exposure - this may be beneficial in some mergers, as there is a lot of detail buried in the dark side of the tree bark an orange leaves. Watch for the sky in the upper areas of the image - it may turn purple-ish - and also watch for haloing around the tree-sky border - an area of abrupt contrast.

Etc.

I will try to provide examples of issues from the various applications I use as a start to the thread. I hope you folks try this out too and add your images and your comments here, along with a brief description of your workflow and how you arrived at your final image. Try re-working the image data too, to see if you can improve the workflow or a choice of setting that may have caused an issue, etc. You do not have to use all of the images in the sets if you do not want to or have to - also, try tone mapping just one of the raw images (the "best" one) and see if you can get a decent result compared to a merged set. Can you control noise and eek out detail without resorting to HDR? Is Luminosity blending or fusion a better approach? How does the new LR/ACR HDR merge to DNG compare - is it robust enough to manage the dynamic range, ghosting, etc.?

Have fun!

kirk

Link to the image data:

Scene01 - 225MB
https://www.dropbox.co​m …g6v59sp/01Sourc​e.zip?dl=0

Scene02 - 82MB
https://www.dropbox.co​m …zrt4du2/02Sourc​e.zip?dl=0

Processed in the SNS-HDR Pro 2.0 Alpha 3. The HDR and single image (_MG_0268.CR2).

Very nice Pebal! I am looking into a virtual machine install on my Mac to try the most recent version of your software. Looks like it does a really nice job with a potentially challenging scene. The support for EXR images makes it even more appealing. Thanks for giving it a go.

kirk

Here is a 40% pedal image, 60% my original blend. As with many HDR images I produce, blending multiple versions seems to work the best, even if the versions go for an overall finished look versus blending highlights in one versus shadows from another. The attached blend is simply placing Pebal's version on a layer above the original and decreasing the opacity to 40% in Normal blend mode.

kirk