Kinmarq is built on open research. Several of the licences below require that we credit the people whose work we use — this page is that credit, and it is kept accurate rather than generous: something is listed here when we actually ship it.
Dog identification datasets
DogFaceNet
Mougeot, G., Li, D., & Jia, S. (2019). A Deep Learning Approach for Dog Face Verification and Recognition. In PRICAI 2019: Trends in Artificial Intelligence, Springer, pp. 418–430. Dataset: doi.org/10.5281/zenodo.12578449 (8,363 images of 1,393 dogs). Code: github.com/GuillaumeMougeot/DogFaceNet.
Licensed CC BY 4.0. We use it to train and validate the dog identification model that powers snout and face matching. Attribution is a condition of that licence.
Machine-learning models
Photographs you upload may be analysed by the following models. Each is used under its own open licence, and none of them is treated as proof of identity — a person confirms every outcome that affects a pet’s record.
- DINOv2 (
facebook/dinov2-small) — visual embeddings. Apache-2.0. - SigLIP 2 (
google/siglip2-base-patch16-224) — visual attributes such as colour and coat. Apache-2.0. - Florence-2 (
microsoft/Florence-2-base-ft) — written description of a photograph. MIT. - DETR (
facebook/detr-resnet-50) — locating the dog in a photograph. Apache-2.0. - Dog breed classifier (
wesleyacheng/dog-breeds-multiclass-image-classification-with-vit) — breed suggestions.
Software
FastAPI, Pydantic, Pillow, NumPy, Transformers, PyTorch and timm (MIT, MIT, MIT-CMU, BSD, Apache-2.0, BSD-style and Apache-2.0 respectively); .NET, ASP.NET Core and Entity Framework Core (MIT); Next.js and React (MIT); PostgreSQL (PostgreSQL Licence); Keycloak (Apache-2.0).
Corrections
If you believe your work is used here and credited wrongly, or not credited at all, tell us and we will fix it.