One important observation about hierarchical organization that I must note. Yes, it is very similar to hierarchy-free keyword grouping. However, it’s just that some search queries are more optimal than others, and that if you want to find something exactly, you have to specify every single keyword value in the search query, then sort them into the proper order to get the filesystem path to the object in question.
Oh yeah, and one important note for searching for objects by their image. Here’s the idea. All you need is a series of pictures to effect a 360 degree view of the object in question. Photos can be 2D rotated for matching as necessary, so all you need to do is take the 2D features out of a new photograph and determine which collection of 2D photographs is best matched by this. And, I note this one, storing only the 3D models of the objects in question is a great way to compress the collection of 2D photographs for the object. However, it does imply a cost in regenerating them, and the 2D search is a rather different method of operation. So, those are the caveats, and it may make sense then to store all 2D photographs pre-rendered for computation. The only time it may make sense to not store them is when you want to crunch your data down to fitting in a small storage space.
Anyways, pretty interesting things going on here. And especially with hardware artificial neural networks, doing image searches by photographing will be really fast and efficient.