AUtomatic generation of (some) metadata?
I use frequently freebies or products from other vendor sites and I am trying to find a way to automatically add metadata to them by inspecting the asset directories.
Clearly defining categories is impossible to do automatically. For instance, finding that a poses should be "By function/Kneeling" or "By Region/Full Body" or a camisole (or whatever name the artist gives) "Wardrobe/Shirt" and not "Wardrobe/Pant".
But some of the ContentType and Compatibility could probably be automatically generated.
I will only consider clothing and poses (that are the most useful for smart content IMHO), but hair (and maybe char) could be processed like clothing.
For poses, one can reasonnably consider that if we go through the directory hierarchy of a given asset starting at People/.../Poses, we can stop once we encounter a dir containing a ".duf" file and that all the ".duf" files below this directory will be poses with a compatibility derived from the path. For instance, if we find in People/Genesis X/Female/Poses/xx/yy/ files named file1.duf, zz/file2.duf and zz/file3.duf, file1.duf, file2.duf and file3.duf may be considered as poses with a compatibility of GenesisX/Female
Ditto for expressions.
For clothing, one could consider that if we go through the file hierarchy similarly starting at People/Genesis X/Female/Clothing, we can stop once we encounter a duf file and that all the ".duf" file in this directory will be "Follower"s with the Genesis X/Female compatibility. It is probably impossible to be more precise for follower and processing materials also is almost impossible. (How to know that a preset applies to the top and not the shoes? or that it is iray or 3DL?)
But with these restrictions, I have the feeling that this kind of information can be generated automatically and that the classification should be successful at least 90% of the time (successful in the sens that, the classification will be very partial of course and some items may be missed, but at least it will be correct for the .duf files that are classified).
What is your opinion? Am I missing some important point? How useful would be this partial generation of metada and what would smart content do with it? Could it be improved?
Comments
The content wizard plugin probably uses that approach for allocating sensible defaults for metadata to items. But with all the vanity folders and artists deviating from conventions by mixing artist names/product names in paths, makes processing of sensible defaults more difficult than it should be.
Oh yes! And it is even worse for freebies! Actually I have a (more or less) working script to convert any asset to DIM which tries to automatically modify part of the hierarchy to renormalize the content, fix typos, etc. It works more or less and this restructuration can partly help with the metadata, but the kind of (dis)organization that you can find is crazy. So I think that one should be very conservative and only do classifications that are almost certainly correct. But if an artist adds materials presets in a "poses" directory no way not to be fooled.