PicAtU: Face Recognition Photo Retrieval

A web application that finds every photo you appear in at an event and lets you download them. Just upload a selfie.

Overview

After a college event, hundreds of photos get scattered across phones and drives. PicAtU indexes an event gallery once, then matches a user’s face against that index so each person can pull out only the photos they are in.

How it works

  1. Enrolment. A selfie is uploaded and converted into a face encoding.
  2. Indexing. Every photo in the event album is scanned with Dlib’s face detector, and each detected face is encoded and stored in MySQL alongside the photo reference.
  3. Matching. The query encoding is compared against the stored encodings, and matches above a distance threshold are returned as the user’s photo set.
  4. Retrieval. Matched photos are shown in a React gallery with a download action.

Architecture

Face detection and encoding run in a C++ layer built on Dlib, compiled with CUDA support so the scanning step stays fast on large albums. The system is split into microservices: a React frontend, a Node.js API, and a Python Flask service that owns the vision pipeline, with MySQL as the shared store.

Tech stack

  • Frontend: React
  • Services: Node.js, Flask
  • Vision: C++, Dlib, CUDA
  • Storage: MySQL

Highlights

  • Attendance photos become searchable by face instead of by filename.
  • Heavy vision work is isolated in its own service and can use the GPU.
  • Modular design keeps the web tier independent of the recognition engine.

PicAtU: Face Recognition Photo Retrieval

undefined

Author

Bibit Kunwar Chhetri

Posted on

2023-08-01

Licensed under