ruvnet/rvface

rvFACE is a face recognition system rewritten in Rust and WebAssembly, based on an existing open source project originally written in Python and PyTorch. It can run natively on your computer or directly in a web browser, using either a graphics card through WebGPU or a regular CPU, and it comes with a web interface for trying it out. The way it works is a pipeline of steps: an image first goes through a detector that finds a face, then a model finds 68 landmark points on that face along with its head pose, then the face is aligned and straightened, and finally a neural network turns the aligned face into a numeric fingerprint that can be compared against another face's fingerprint to produce a similarity score, with a score above 75 counting as a match. The project is organized into several Rust pieces including a core math library, the machine learning models, a command line tool for detecting and comparing faces, and browser bindings, alongside a separate web app built with Vite and TypeScript. According to the README, the project is complete: all three neural networks have been ported and checked against the original PyTorch versions with very close matching results, the full pipeline reproduces the original demo's verdict on its test images, and the browser version runs the same engine using WebAssembly with automatic fallback from WebGPU to CPU. It includes 66 Rust tests covering math, model parity, and full pipeline runs, plus published benchmarks. The code itself is MIT licensed. Two of the model weight files are properly licensed and included in the repository so the browser demo works out of the box, while a landmark model with no clear upstream license is fetched separately rather than redistributed. The README is explicit that this is biometric processing software meant for consent based uses like authentication, personal photo tools, or research, and warns against using it for surveillance or identifying people without their consent, noting that laws like GDPR and BIPA may apply.