Hugin

Running multiple Nona and enblend in parallel

Asked by Faveo Hörold

Hi,

Thank you for the amazing software suite that constitutes Hugin!

I've recently shot some very large panoramas and it's no longer feasible to stitch them on my personal machine due to limited memory (500 Megapixels on 16GB RAM). Luckily I have access to a machine with 128 cores and ~500 GB of memory. I expected stitching to utilize the extra cores efficiently, but it's no faster than my laptop. Of course, my problem of limited memory is solved, but I'd like to see if I can speed up stitching nonetheless.

My understanding is that nona and enblend are multi-threaded, but work on one image at a time. I will assume that these tools do not scale past more than a few 10s of threads. On the other hand, with this amount of memory it should be feasible to remap all images at the same time; each on an individual thread. I guess enblend would need to work with some kind of binary-tree strategy, always blending pairs of images.

Is it possible to call the individual stitching steps and tools manually from the command line? Then I would write a bash script that splits up the work between tools with something like gnu parallel.
I would like to avoid futzing around with the batch processor internals, but if there's no other way, I'm also happy to do some light coding.

Thanks for your help!
Faveo

Question information

Language:
English Edit question
Status:
Answered
For:
Hugin Edit question
Assignee:
No assignee Edit question
Last query:
Last reply:
Revision history for this message
tmodes (tmodes) said : 		#1

When processing several images in parallel the hard disc becomes the bottleneck. You need to read and write big images in parallel - then the different threads will blocking each other.

Concerning enblend, you can`t take simple 2 images and process them. You need to check that you blend overlapping image pairs first and then blend the image pairs further. So it's not a simple splitting, you need also to synchronize/queue the blending tree. But in this case the hard disc becomes also the bottleneck.

Can you help with this problem?

Provide an answer of your own, or ask Faveo Hörold for more information if necessary.

To post a message you must log in.