An issue that strikes me as I look at the pots is that quite a lot of the pics are taken at an angle.
Now, if the angle is extreme (the photo is of the bottom of the pot), then it's useless. But there are many like this:
But I can't access it. Never mind: it strikes me that our registration automatically compensates for this. Suppose we were to take, for some class of pots, the dead-on view as the gold standard and register all other pots to that. The resulting transformed shape should tell us what its dimensions really are. Then you compare different classes of pots.
This does assume that the differences within a class of pots is due to differences in photographed angle. So you can't get an estimate of variance within a given class of pots. But it is worth thinking about whether we can use slightly angled pots rather than just chucking them.
An issue that strikes me as I look at the pots is that quite a lot of the pics are taken at an angle. Now, if the angle is extreme (the photo is of the bottom of the pot), then it's useless. But there are many like this:
/Users/Armand/Dropbox/Greek Vases images/original images/edited_images/good beazley images_classified/good_corinthian/corinthian_aryballos_NA/pot_number525005.0_url.img_3.jpg
Not quite dead on, but the whole shape is there. Now there's a paper that gives a method to compensate for this. https://ieeexplore.ieee.org/document/1309275
But I can't access it. Never mind: it strikes me that our registration automatically compensates for this. Suppose we were to take, for some class of pots, the dead-on view as the gold standard and register all other pots to that. The resulting transformed shape should tell us what its dimensions really are. Then you compare different classes of pots.
This does assume that the differences within a class of pots is due to differences in photographed angle. So you can't get an estimate of variance within a given class of pots. But it is worth thinking about whether we can use slightly angled pots rather than just chucking them.