urLifePrint
enrollment + matching live · 2026-08-15We can recognise your dog from a photo, and we will tell you when we are not sure.
The interesting part is not the recognition. Every recognition system can return its closest row and call it an answer. This one returns NO MATCH instead — proven on a dog it had never seen, where the nearest enrolled animal scored 0.15 against a 0.50 bar. It declined rather than guessed, and nobody was handed the wrong dog.
Implausibility raises the bar rather than closing the door. A candidate whose recorded size or breed disagrees with the photo must clear a correspondingly higher threshold — but the penalty is capped below the headroom, by construction, so a cross-country dog with overwhelming image agreement can still surface. A filter that could eliminate a candidate is a filter that can lose somebody’s animal.
In an offline test on 210 dogs the model had never seen, it picked the right dog first 97.6% of the time from a four-photo enrollment, and made a confident match in 80.6% of searches while holding false matches under 1%. How it performs on phone photos taken in real conditions has not been measured yet.
