Can you clarify a better workflow for bracketed exposure HDR 360-panoramas?

Question

After reading the thread about HDR Panorama workflow I was a bit confused. I need some help with clarification. In my current workflow I have to edit the nadir shot(s) using The GIMP. I do my best to take out the tripod and pano-head by rotating Nadir1 on top of Nadir2 and mask out what I can. This become very complicated if I take exposure bracketed shots and attempt to build what I believe is called an HDR Panorama.

I think my confusion comes from my limited understanding of what bracketed shots are and how to use them. For example: I don't understand why there are so many HDR programs (ie luminescence) when Hugin is there for just such a stacked project. However that idea did not pan out either.

First what is the difference between a bracketed exposure stack and an HDR image? Also if editing the nadir in order to stitch I have to tone map and if using The GIMP downgrade to an 8-bit image losing all the data that Hugin would have needed to have for best alignment and exposure blending.

So if I was making just a simple HDR photo or horizontal panorama either HDR first or stitch first works however if you need to edit the Nadir shot prior to stitching (as needed in a 360 panorama) then you run into the chicken before the egg problem.

What knowledge am I missing? How do you reconcile a good work flow to make a well tone mapped HDR 360 panorama with edited out tripod from bracketed exposure shot in raw format using tools like Hugin, The GIMP, etc.?

EDIT: After reading a little further into the above question and this question I realize that I'm not asking a direct question here but clarification into a specific workflow that uses the workflows described in the other questions together. Also by HDR I probably mean Exposure Fusion instead. If anyone is willing to clarify my confusion I would be grateful. I did learn quite a bit from the other questions as well.

Answer 1

Yes, that's a tough problem – you have the option of

1. Editing the three bracketed nadir shots separately in GIMP (bad idea, because you'll never be able to rubber-stamp exactly the same pattern in each exposure).

2. Creating tone-mapped stacks of each individual tile, then editing the nadir shots in GIMP and stitching them in Hugin (not recommended, especially if you're using tone mapping operators such as Mantiuk06 and Fattal – they will take matters into their own hands and really muck around with the exposure, particularly at the edges of the image, so when you go to stitch the image, you'll get all sorts of patches with dissimilar brightness).

3. Stitching all the images (including nadir shots) in Hugin, leaving the nadir shots un-edited, saving it as a low-dynamic-range (8-bit) image, and then editing out the tripod in GIMP. Depending on the projection you use, it might be quite difficult to edit out the tripod – think about how huge and stretched Antarctica is in most map projections! If you're doing one of those 'little planet' projections, though, it should be no trouble at all.

With the open-source tools at your disposal, I'd go with option 3. Unfortunately, every time you output a new version of that pano from Hugin, you'll have to take the 8-bit image and edit out the tripod. Kind of annoying, but it will allow you to maintain the HDR integrity through most of the workflow. Ideally, you want to tone-map it as late as you can.

In Hugin parlance, 'high dynamic range merged stacks' means an HDR image – or rather, each bracketed stack (three low-dynamic range images) is exported as a separate HDR tile.

In Hugin, 'exposure fusion' creates a low-dynamic-range tone-mapped image using an HDR image (that Hugin creates and then throws away). Hugin's tone-mapping is pretty good, although I find it creates pretty low-contrast images.

The purpose of Luminance in this workflow would be to play around with different methods of tone-mapping. I often output the stitched pano as an HDR image (in EXR format) from Hugin, sidestepping Hugin's tone-mapper, and then play around in Luminance. Landscapes can have a lot of contrast between highlights and shadows, and it sometimes takes a bit of effort to get things balanced. My favourite method is:

1. Tone-map the image with Reinhard or Mantiuk08, and save that JPEG. This will give you a low-contrast but realistic base image.

2. Tone-map the image again with Mantiuk06 (contrast 0.2, saturation 1.2) or Fattal and save that JPEG. This will give you an unrealistic image that has strong local contrast (details are enhanced) but low overall contrast (shadows are lightened, highlights are darkened).

3. Load up both images as separate layers of the same image in GIMP (image 1 on the bottom) and play around with the transparency on the second layer until you get a pleasant looking image. You might want to set the second image's blending mode to Overlay and knock the opacity back to 25% for starters.

Answer 2

One issue that you may run into with multiple exposures from bracketing is that the photos from each set do not line up properly. However, it will be smoothest to create the panorama prior to HDR since it will combine the photos using the originals.

I know with Photoshop, you can create awesome HDR photos without any extensions and it will auto align your three bracket shots panoramas for you.

If you have the extra money, I would go that route. I've used Gimp plenty of times, but I always revert back to Photoshop.

Answer 3

I think your problem is that you want to nadir patch before you stitch, rather than after. I never do it before. I always do it after. So, if I'm exposure fusing a bunch of bracketed sets as stacks in Hugin/PTGui, I always do the stitch first. Both Hugin and PTGui now have masking features, which will let you eliminate the panohead and tripod from the final stitch.

You'll have "holes" in the final pano if you aren't going to patch with handheld nadir shots where the tripod and panohead are removed, but if you choose a featureless place to set the tripod down, simple cloning, patching, or content-aware fill can typically do all you need, if you remap the pano so the nadir's in the center (a simple pitch adjustment), or map the floor out to a cube face with something like Pano2VR.

I also exposure fuse rather than HDR, because it eliminates the entire need for the weird-format HDR file, as well as tonemapping, and the default typically looks more realistic than with HDR, since enfuse is simply selecting pixel values from the member images based upon exposure and saturation (sidenote: if you decide to weigh contrast more heavily in the three selection criteria, you can also use enfuse to focus stack).