DogArmor
computer-vision
pytorch
python
Identify a dog’s breed from one photo, estimate its measurements, and recommend a harness, collar and leash that fit.
Overview
Upload one photo of your dog and DogArmor identifies the breed, estimates its girths and back length, and recommends gear that fits from a 36-product catalog.
Breed classifier
- An ImageNet-pretrained EfficientNetV2-S reaches 97% top-3 and 81% exact breed accuracy on 500 held-out photos.
- Exported to ONNX and run with ONNX Runtime, it predicts in about 20 ms per photo on CPU.
From breed to measurements
DogArmor estimates girths and back length for 118 breeds from a single photo. It scales each breed’s weight range by the cube root of weight, then adjusts for build.
Recommendations
A 13-question profile and a budget are scored against the catalog to pick a harness, collar and leash. Safety rules are built in. For example, it never recommends head halters for flat-faced breeds.
Shipping it
A pytest suite runs in CI, and every push to main auto-deploys to Railway with Docker.
Stack
Python · PyTorch · ONNX Runtime · Docker · Tailwind CSS · Railway