I noticed the merging partial reconstructions feature could easily introduce noises to the model.
My customer usually collects both nadir images and oblique images, sometimes they didn’t do a good job to have enough overlapping. Previously, only the partial reconstruction for nadir images will be used so at least the orthophoto and dem will still be good enough. Merging the partial reconstruction for oblique images could make two misaligned reconstructions into one and hence bad final results. And I have to manually remove all the oblique images to reprocess.
Maybe we should add a flag to disable/enable this feature?
I think a script to filter out oblique images might be the way to go here, based on the XMP/EXIF information; partial reconstructions indicate a dataset with insufficient overlap (bad input). Note it doesn’t necessarily mean that nadir images end up in one reconstruction and oblique images end up in another. They could be mixed and matched. So disabling the merge might not necessarily improve the situation (it might in some cases, but not all).
This could be a good candidate for a contrib script (help us write one?).
Oblique images are collected intensionally as they can be helpful to reconstruct better texture for some vertical surfaces in the 3D model. So if they can be matched together I would still prefer to keep them.
For example, sometimes the red cameras are not collected so the oblique ones can’t be matched with the nadir ones. And hence 2 partial reconstructions could be merged with some misalignment, which introduces noise. If using the first reconstruction only, I can still have good orthophoto and dsm for this.
I understand that this indicates bad overlapping and how the data is collected is not a typical mapping survey flight. This should be the user’s job to provide proper input. I just hope to be able to control the behavior when such reconstructions happen.
(I had a script to simply filter the pitch attribute in the Photo object to detect the oblique images to serve other purposes, but only limited to cameras that collect such info)