Dear Maddy,
Glad to hear other people are trying similar projects! If you want advice from someone with a lot of experience working with legacy data, I believe user Wishgranter (see above) is a good connection: he's worked on similar projects in a cultural heritage context and has helped a lot with this project. Since you're also a maritime archaeologist, David Cockey (again see above) is similarly super friendly and helpful, and has loads of experience working with modeling of historic vessels! With all this cultural heritage / maritime archaeology interest going to PhotoScan it might almost be worth setting up our own little user group?
As to your question, I have not had access to the original dias so no rescans were made BUT the two major "breakthroughs" (which I apparently failed to update here) were:
a) realization that
pre-processing in Photoshop (or similar software) is very useful! Since my photos contained a lot of vignetting, with excessive highlights in the center and shadows at the borders of the picture, the main tool I used was Image > Adjustments > Shadows/Highlights. Additionally I did some sharpening using Filter > Sharpen > Unsharp Mask. After trying a few samples, all photos were processed together in Batch Process.
b) tip by Wishgranter to
resize all photos to the same size: all pictures were scanned to different sizes, which is why PhotoScan attempted to calibrate them all individually, leading nowhere. However, by adding a simple black frame around each image I resized all images to slightly bigger than the biggest picture, and thus PhotoScan calibrated them all together, while ignoring the black border since no matches were found in that zone. Image resize and black border were again done in Photoshop with Batch Process.
=> example of initial picture
http://www.mediafire.com/view/g42w0z0e6g6sv11/111-006N.jpg vs final output
http://www.mediafire.com/view/7657j73883gotda/111-006N%20edited.jpg .
I've been busy with other stuff so I don't yet have a full 3D model of the site, but this methodology has enabled me to process at least very large chunks of the site at once, with nice results (where before I really got nowhere). There's some problems of insufficient overlap at certain points which of course can't be overcome by simple pre-processing but I'm quite confident that the final result will be good (perhaps using some manual chunk alignment). I'll try and remember to post an update once the whole set has been processed!
Hope this helps,
Cheers,
Tom