// The explanatory blurbs under each tool. Shared by the home page and the sample-dog profile so the // wording (and the quoted accuracy figures) can only ever be changed in one place. /** Under "Image matching": how the search works and how to read the scores. */ export function MatchingNote() { return (

How it works: A model turns each image into a feature vector (a list of numbers) describing the dog. This is used like a fingerprint. Visually similar dogs have similar features, so comparing features against the found-dog database enables the tool to rank the closest look-alikes. The top 10 matches are always returned, and the higher the score, the closer the match. If a dog is in the database, it is usually returned as the top match or, at worst, should be among the top 10 matches over 95% of the time. If all the scores are low, the dog is likely not in the database.

); } /** Under "Breed estimation": what it predicts, how to read the scores, and that it is separate. */ export function BreedNote() { return ( <>

How it works: The breed classification model attempts to predict a dog breed from an image. The higher the score, the more confident the model is about a specific breed prediction. The top 10 breed guesses are always shown even if most of the scores are very low. This does not imply a mixed breed dog. Many breeds have similar visual features and photo angles vary, so the correct breed isn’t always ranked first. However, for clear, known breed photos the correct breed should be listed in the top 10 over 95% of the time. We don’t have accuracy stats on mixed breed images, but the model will certainly surface breeds with similar features.

Note: The breed classification model is entirely separate from image matching and is included here mainly for fun and interest.

); }