I am currently implementing a similarity check for uploaded images against our Cloud Recognition Target Collection using the Similarity Check endpoint as described in the Manager API docs. It documents that a value of between 0 and 1 is returned for the similarity, the higher the number, the more similar they are. For the purposes of recognition, what should be the cutoff to "reject" a new upload. Should a new image that's > 0.5 be rejected because it's too similar and there will be recognition conflicts? Obviously 1's should be rejected but I am unclear on how similar is too similar/where it will start causing recognition conflicts.
Only a high value of similarity can be a problem for recognition. You can expect problems when adding a new image on the database which has a similarity score above 0.8.