Hi Alexey,
Thanks for your response but I think that you have misunderstood the question.
The option is always checked.
An initial alignment of a subset of the images is created.
Resetting and realigned those that failed to align doesn't make a difference,
so I remove them and then re-import them to initiate the incremental alignment.
Now, this sometimes works fine (new features are detected and the re-imported images are aligned with the rest)
but sometimes it keeps the alignment of the first subset, but removes their sparse point cloud and the new images have detected features which are not matched.
The result is the first subset aligned (without sparse pcloud) and the additional subset not aligned with detected features but no matches, not even between them.
When I align the first subset alone and then add the second one, it usually works.
However, the points is not to do this twice so that we can figure out which subset will fail and process it sequentially.
regards
Please also consider the 'Duplicate chunk' option for the key points
http://www.agisoft.com/forum/index.php?topic=8265.0