Asking for a friend – my wife actually 
She has been slowly getting rid of duplicates in Photos. However, since upgrading to Tahoe, as well as offering obvious pairs, it occasionally offers up to six photos as ‘duplicates’ which are completely different – different subjects, different locations, taken on different dates, even on different cameras.
Can anyone throw any light on the algorithm which would class such disparate photos as ‘duplicates’? Is this a ‘feature’ of Tahoe? Anyone else noticed anything similar?
My comments are based on extensive observations of Photos. Your (wife’s) observations are an additional data point. My questions are intended to clarify the data.
My guess is this is neither bug nor feature, Photos is still “thinking about it’. Or perhaps a minor bug, in that yet well-defined results are reported. Is people and Pets still “Finding people”?
How long since the upgrade to Tahoe? With each System upgrade Photos changes, including its image sorting and identification algorithms, as you suspect. In general, one type of change is to include more details in the analysis. More details means the calculation has become more complex, and the previous results need refinement. Without knowing the details of Photos’ process, we do know identified and sorted photos is the result of its iterative refinement. This takes a lot of calculations, which takes a long time.
If she lets Photos process for a few days (Photos running in the background, plugged in to power), does the number of duplicates found, correctly or incorrectly, change?
Was Tahoe installed over an earlier system, or is this a new machine? Photos’ analysis of your photos is iterative refinement of data local to your machine, which produces a predictive model which is also local. How previously calculated results are carried through to new software or hardware is not clear to me, but I think has interesting potential. For example, a clean install I suspect would eliminate the predictive model which resulted from all previous analysis of the photos library, but not the resulting assignment of names, etc., to elements of each photo.
Thanks for your comments. Tahoe was installed as soon as it was available and was installed over the previous system. The way Denise (my wife) has been working, she will launch Photos when she has a spare minute and the click Duplicates to reduce the volume of them. For a variety of reasons she had lots of genuine duplicates. I’ll attach screen-shot of a set which was offered to her as duplicates.
As you can imagine, she was genuinely puzzled to be offered such a diverse set of images. I hope it provides a useful data point for your further study. Thanks again for taking the time to reply.
