Face Recognition Starter for ZETIC Melange

This repo now contains a Python helper that downloads a pretrained ONNX-based face recognition package, runs it locally, and points you at the ONNX file you can upload to ZETIC Melange.

What this does

• Downloads an InsightFace model pack into the repo.
• Runs face detection plus face embedding extraction locally on CPU.
• Lets you build a face database from input_images/<person_name>/....
• Lets you match a new image against the enrolled database.
• Lets you run live webcam recognition with green known-face boxes and red unknown-face boxes.

Important Melange note

ZETIC Melange itself runs models through the Android or iOS SDK after you upload an ONNX or .pt2 model to the Melange dashboard. This Python file is the local prep and validation layer. It helps you get a facial recognition model running and identify the ONNX asset to upload.

Setup

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

Usage

Prepare the model package:

.venv/bin/python face_recognition_melange.py prepare

The default model size is medium, which maps to buffalo_s. You can choose the size directly:

.venv/bin/python face_recognition_melange.py prepare --model-size small

The size mapping is:

small  -> buffalo_sc
medium -> buffalo_s
large  -> buffalo_l

You can still choose the pack explicitly if you want:

.venv/bin/python face_recognition_melange.py prepare --model-pack buffalo_s

Enroll a person:

.venv/bin/python face_recognition_melange.py enroll alice /path/to/alice.jpg

Build the full database from person folders:

.venv/bin/python face_recognition_melange.py build-db --image-root input_images

Match a new image against the local database:

.venv/bin/python face_recognition_melange.py match /path/to/query.jpg

Directly compare two images:

.venv/bin/python face_recognition_melange.py compare /path/to/a.jpg /path/to/b.jpg

Capture input images from your webcam:

.venv/bin/python capture_face_inputs.py --person alice --prefix sample --count 3

This saves images like:

input_images/alice/sample_01.jpg
input_images/alice/sample_02.jpg

Run live webcam recognition:

.venv/bin/python live_face_recognition.py --db face_db.json

Model source

The script downloads official InsightFace release packages such as buffalo_s.zip, buffalo_l.zip, or buffalo_sc.zip from the deepinsight/insightface GitHub release URL.

Licensing

InsightFace's repository notes that its pretrained face recognition models may have separate licensing constraints. Check the current InsightFace license terms before using the pretrained model in production or commercial settings.