Good dog.
Great mystery.
Let’s sniff it out.
Those ears. That face. The unmistakable personality.
Get to know the breed behind your best friend, one photo at a time.
Your dog. Our best guess.
Introduce your dog
PHOTO UPLOADYour photo is used for this prediction, then discarded.
Meet the possibilities
TOP 5 MATCHESA familiar face.
A new discovery.
Add a photo and we’ll fetch the five breeds
that look most like your dog.
Following our nose…
Looking for familiar ears, faces, and features.
A case of mistaken
dog-dentity.
A curious guess, not a pedigree. PetSnap compares appearances across 120 breeds; it doesn’t determine ancestry or mixed-breed percentages.
How does a photo become a prediction?
Before you arrived, PetSnap practiced on dog photos labeled by breed. It learned visual patterns that help tell breeds apart. First, a separate detector checks the full photo for a dog. If it finds one, the breed model takes a closer look. The detector can miss dogs, so a clear, close photo helps.
-
01 / THE INPUT
A photo becomes numbers.
We resize your photo and crop its center. The colors in that small square become numbers the model can work with.
-
02 / THE LEARNED PATTERNS
Little clues come together.
Layers of the model combine simple patterns, like edges and textures, into more complex ones that can help distinguish one breed from another.
-
03 / THE PREDICTION
The closest matches rise to the top.
The model gives each of its 120 breeds a score. We show the five highest. A higher score means the model favors that breed—it doesn’t guarantee it’s right.
Learning happens during training. Uploading a photo runs the trained model; it doesn’t teach PetSnap a new breed or add your photo to the training dataset.
Simplified illustrations, not a visualization of your photo’s actual prediction.Dog person.
Data person.
Why not both?
PetSnap is an independent machine learning project exploring what a neural network can learn about our four-legged friends.
More to discover in the full appThe full PetSnap app will feature American Kennel Club information for each dog breed, so you can get to know more than just a name.
Explore the project on GitHubPretrained on ImageNet. Fine-tuned for dogs.
A little more about the model +
Trained on the Stanford Dogs dataset with PyTorch. A photo is resized and center-cropped before the model returns five ranked breed predictions. Scores reflect the model’s output, not a guarantee of a correct identification.
The example photos come from the training dataset and are for demonstration, not an independent accuracy test. Unfamiliar breeds, mixed breeds, and non-dog images can produce confident but incorrect results.