I presume the preselector decides which images to match against which other images. I would like every possible pair of images to be matched. Is that possible?
Hi Aaron Curtis
Preselector is related to point selection for actual alignment. its useles to do every pair image matching as it not work as you describe. We set a bit strict the alignment values but its great to learn from that with very fast feedback. So if get not proper alignment can can see multiple COMPONENTS then try inspect the reasons for this behavior.
Can post some of you results so can see cameras in scene ?
Hm, I had some trouble understanding your reply, probably because the preselector is not what I think it is (there is a similar-sounding feature in Agisoft Photoscan and I assumed it was the same idea.)
The scene I’m currently processing will be done in about 2 hours; I’ll post about it then. It is a cave with 2000 photos. Most of the photos are not very good (some are out of focus, some have bad lighting). The photos are very dense, though, so I hope I can reconstruct the whole cave.
Two more questions in the meantime, though:
- Do the photo filenames or the sequence in which the photos were taken matter at all?
- Is there a way to display two components in 3D at the same time in different frames of the window?
(By the way, the more I learn this software, the more amazed I am by it. Some really innovative stuff.)
Hi Aaron Curtis
So get the difference 
Two more questions in the meantime, though:
- Do the photo filenames or the sequence in which the photos were taken matter at all?
- Is there a way to display two components in 3D at the same time in different frames of the window?
No, filenames matter not for actual alignment…
No way for now to display 2 components in 3D, take again a look on the video how im work with it. As im say, when get it in hands will not need to have this “feature”. Its about experience and can say in few days get it in hands and will not need to see 2 components at all… This was used very heavy on State level heritage project with 100´s of BIG dataset and we set it from customer feedback this way…
Look on very useful info every CP have own projection “error” so can easy find out where the CPs are not placed properly ( per COMPONENT )
About the multiple components thing, the reason I want it is because I use the shape of cave passages to figure out the rough relative position of my components I can find where to add control points. See the two components attached, which are different walls of the same cave passage. I think the curves in the passage line up, but I need to be able to see them side-by-side to be sure.
Unlike your heritage projects, there are no features in the cave which are easily identifiable to use as a control point. I probably should have put markers out during the photography.
Another thing you can do is select all of the cameras in the 3D view and then look at the highlighted images in the image list. I find this helps locating images to add CPs in. I shoot long paths similar to your cave scenario and have found that the image sequence can make a difference. If I return to one area to add more detail it can help to reorder the images so that these appear next to other images of the same area rather than at the end of the entire sequence.
Hi Ben, thanks for the tip. Clearly for a big project like this it’s important to know what photos are nearby each other like you say but unfortunately I seem to have misplaced my field notes! When you say reorder images… How do you reorder them? Do you mean you rename them so that the order changes?
I finally understand the (huge) difference between RC’s “preselector” and Photoscan’s “pair preselection.” They are totally different things. Wishgranter was trying to explain this to me earlier in this thread and I wasn’t getting it, sorry!
In PhotoScan, the preselector determines which images will be matched against each other. In RC, however, all images (at least, if “enabled” and not “locked” or in a “pose group”) are matched against all other images. The preselector, instead of choosing images, chooses the best features to use when matching those pairs. (How does it determine which ones are the best?)
PhotoScan automatically discards features that don’t agree with the rest of the matches (they are shown in red if you go to the PhotoScan matches view). That is the part of PhotoScan which is most similar to RC’s preselector.
Aaron Curtis wrote:
Hi Ben, thanks for the tip. Clearly for a big project like this it’s important to know what photos are nearby each other like you say but unfortunately I seem to have misplaced my field notes! When you say reorder images… How do you reorder them? Do you mean you rename them so that the order changes?
Yes. I use Adobe Bridge to manually sort the images by similarity and then run a Batch Rename. If you need a reference to the original filename you can save this into the metadata as part of the renaming step.
On the topic of caves… A series of shots with a wide angle lens (or it would be great with a full fisheye lens distortion model ;)) pointing along the cave could be useful for anchoring your other shots to e.g. https://photosynth.net/preview/view/e35f3cb4-bdff-44e1-9a8d-cc3bccb13a7b (not mine). I’ve done a reconstruction of this image in both Photoscan and RC and it works pretty well. It still needs extra images, but it provides a contniuous model on which to add them