Preselect image pairs while aligning chunks

[superpulpo@gmail.com]

Hello,

I have a project with 1600 photos I'm processing in 2 chunks of 800 each chunk. The photos are in a single row linear transect of a coral reef. I'm processing up to the mesh, then aligning and merging the meshes to generate a single Orthomosiac of the entire scan. Aligning the chunks takes a very long time, and I would like to speed up the process. As the scan is a single row of images, only 20 or so of the images have overlap. If I could tell Metashape which images have overlap it should speed up the alignment time dramatically.

I checked the manual and found this:

Preselect image pairs (Point based alignment only)
The alignment process of many chunks may take a long time. A significant portion of this time is spent for matching of detected features across the photos. Image pair preselection option can speed up this process by selection of a subset of image pairs to be matched.

But I can't figure out how to actually preselect the image pairs for alignment. No matter what I try the align chunks process seems to churn through the whole 1600 photos. Does this feature even do what I want? How do I use it?

[JMR]

Hola superpulpo:
in your case I think it would be faster to align all photos at onece selecting sequential strategy for preselection. If you have about twenty redundancy that is to say 95% front overlap, I would say you can apply an important overlap reduction either by blindly sampling photos in advance, or by a more precise method, after alignment, apply overlap reduction based on a rough mesh that you can obtain from the sparse point cloud.
Another approach could be to let some photos to be included in both chunks that you want to link, that will make chunk to chunk alignment "by cameras" almost instantaneous.
We have assisted some users doing the same type of surveys in Florida. If you want to know more, please contact my partner Simon Brown. https://accupixel.co.uk/2022/05/13/scripting-the-value/
Regards

José Martínez
Accupixel & Geobit
Metashape training and consultancy