Any reason to use Enblend over Verdandi?

[Monkey]

I haven't used Hugin for a while, but I recently started updatng my old Multiblend project as I had recently rewritten the pyramid processing part of it for another project.

I hadn't used the built-in Hugin blender (Verdandi), so I gave it a try and was pleasantly surprised with the results. Wtih seam=blend, it does a great job of matching the exposure of images (even if it does so relative to the first image in the list, seemingly undoing much of the work Hugin may have done in balance the exposure of control points), the end result looks very good, and seems to do just as good a job at optimising seams as Enblend does.

With that in mind, and given that Verdandi must be at least dozens of times faster then Enblend (and likely gets exponentially faster as mosaic size increases), what are the reasons, if any, for using Enblend (and still having it be the default blender in Hugin) over Verdandi?

[David W. Jones]

Only time I tried Verdandi, it didn't do as nearly a good job as enblend.

--
David W. Jones
gnome...@gmail.com
wandering the landscape of god
http://dancingtreefrog.com

Sent from my Android device with F/LOSS K-9 Mail.

[Monkey]

Did you try both seam modes? They do quite different things, and the default is not the Enblend-like one.

[David W. Jones]

Haven't tried it for years. Verdandi's default blend wasn't even close
enough to quality to use.

Enblend has been working fine ever since and its speed is fine with me.

My password is the last 8 digits of π.

[T. Modes]

Hi Monkey,

verdandi has 2 blending modes:

- hard seam: this is the fastest one and relies to 100 % on Hugin photometric optimization. For me this works fine in about 3/4 of the cases.

If there are bigger color/exposure differences at the image edges between the images then I use blend seam (or enblend)

- blend seam: it uses the same seam finder as in the hard seam case. The remaining differences between the images are blended with a Poisson equation solver - something completely different from enblends pyramids approach. This algorithm can also blend away some higher color shifts between images. But it still benefits from Hugin photometric optimization - especially when the photometric optimizer has corrected the vignetting (color and exposure is not so important in this case).

But I got some reports where verdandi blend produced bad results. This happens mostly when using bigger images. This issues are not there when reducing the output size. So it is very difficult to debug in the case. Fixes for this are always welcome ;-)

Because of this issue and for backward compatibility I have hesitated to change the default. But each user can change the default blender in the preferences.

PS:

> Hugin may have done in balance the exposure of control points

The control points are only used for geometric optimization. They have no effect for the photometric optimizer. Here only the current arrangement of the images is used for probing the photometric differences between the images.

[Monkey]

Thanks for the info. I wondered if a combination mode could be made, which corrects the colours using the Poisson equation (something for me to learn about - got any links? Is it a frequency domain thing?), but still uses a hard seam to connect the images without trying to blend any features (which blend mode appears to do, as I get smears very similar to Enblend/Multiblend when I use deliberately misaligned images).

I've been updating Muliblend and while it is faster than both Verdandi and Enblend, I haven't yet come up with a reliable way to avoid overblending (Enblend uses different numbers of levels for every image, as the overlap allows, but as well raising the possibility of inconsistent blending this also seems to result in later images being blended more than earlier ones).

[T. Modes]

As I already written verdandi --seam=blend is already using a hard seam as first step. Then the already seamed images are used for Poisson blending. Here only information from the direct neighbourhood of the seam of the first image are used as border condition. So when the differences at the seam are too big you will also get smearing with this algorithm - at least this is my experience.

You will also get some black smearing when the already mentioned bug appears.

Maybe it is better to combine the effort and work on one implementation instead of writing 2 different programs.

[David W. Jones]

Well, it seems to me, the only advantage of verdandi over enblend is speed?

--
David W. Jones
gnome...@gmail.com
wandering the landscape of god
http://dancingtreefrog.com