Mike Buckley wrote:Mike Buckley wrote:do you have the "Also assign its parents" option enabled in your labels?
No. I also don't embed synonyms in the image file.
Synonyms have no effect on the number of labels anyway - but they indeed may affect the number of final keywords. As a matter of fact, it would be interesting to run a similar script on the number of final keywords in the metadata (which may be larger due to synonyms and to enabling "Include all parent level labels as keywords").
You're correct that I use the David Reicks controlled vocabulary. The reason I'm so compulsive about assigning catalog labels is that I enjoy making theme-based slide shows.
That's a very good reason.
As an example, If I want to show photos of circular red objects, I can easily make that happen, not that I've ever done that.
Actually, inviting family & friends to show them a circular-red-themed slideshow seems like a very cool (and artsy) idea. Believe me: I'm only half-joking here!
Andrey_Ra wrote:I'm still waiting for more reports. vlad, how about to participate in survey?
Hi Andrey, thanks for the invitation - but I'm going to decline for now, since my current results would not be useful (even to me):
1) I have many imported images which are not yet cataloged at all.
2) I don't (extensively) label all the images that I import, review and archive - only certain selections (collections). (I know, that's weird.)
3) I'm considering some labelling strategy changes (aka: forever experimenting )
This said, let me tell you kudos for writing such a cool script and promptly updating the graphs, it's interesting to look at them! From what I gather, the label number distribution for most people is close to normal (gaussian) and peaking somewhere around 3 to 6 labels, followed by a long tail to the right. (Actually, I think that's called a log-normal distribution.) It's funny, because the only exception so far seems to be Hert, which has a completely different graph: it shows a logarithmic distribution, with a vast majority of single-label images and then exponentially decreasing percentages as the number of labels gets higher. Just wondering: is this a case in point of less is more? (I largely agree with this principle myself - I just find it hard to apply it in practice! )
A diagram with a single synthetic graph, displaying the overall label number distribution based on all the individual distributions (percentages), would be very useful too!