What is difference?

[yeskent85]

What's the difference between fiftyone.brain.similarity.find_unique() and fiftyone.brain.compute_uniqueness()?
The output of both the functions using the same embeddings are slightly different.

[swheaton]

The difference is slightly subtle.
compute_uniqueness() adds a uniqueness score to each sample representing how unique it is with respect to the rest of the samples. Taking the top N would give you the N samples that are most unique compared to the entirety of the dataset.

find_unique() selects count samples that are maximally unique with respect to each other.

To explain with help of a degenerate case ... let's say we have 1000 images of dogs and 2 duplicate images of a tree. Those 2 tree images are probably going to have very high uniqueness scores because they're very different than the 1000 dog images. But if we call find_unique() then you would not expect to get the 2 tree images as a result because they are not unique with respect to each other.