f2 — Face Detector (ONNX)

Lightweight face detector (f2.onnx) used to localize faces in the PULAO vision pipeline (faces feed the ArcFace embedding model).

⚠️ Provenance / license — UNVERIFIED. The origin of this model is unknown. It is a 2.4 MB detector exported to ONNX from PyTorch; the exact architecture, training data, and license could not be determined from the file. This repository is kept private until provenance is confirmed. Do not redistribute publicly without first establishing the license.

Files

• f2.onnx — 2.4 MB
• SHA-256: 15bcd6208f8959b28d07fd8488ec011368d666421eda9674d248f1d48d869ada

Inputs / outputs

• Input input: float32 [1, 3, 320, 320] (letterboxed square, scaled 1/255, RGB in this pipeline).
• Output: face-detection tensor; the pipeline reads [:, :4] boxes (xywh) and [:, 4] scores, then applies NMS.

Usage (onnxruntime)

import cv2, onnxruntime as ort
sess = ort.InferenceSession("f2.onnx", providers=["CPUExecutionProvider"])
blob = cv2.dnn.blobFromImage(img, 1/255, (320, 320), swapRB=True, crop=False)
out = sess.run(None, {"input": blob})[0]

Intended use

Face localization feeding ArcFace embedding. Research / prototype only.