YuNet
A 228KB millisecond-level face detector from the OpenCV Zoo — five landmarks included, fast enough for any edge device.
About
YuNet grew out of libfacedetection, the long-running tiny-detector project by Shiqi Yu. It is a purpose-built tiny CNN — about 83 thousand parameters — that outputs bounding boxes plus five facial landmarks, and it runs in milliseconds on a CPU. OpenCV ships it in the OpenCV Zoo with an official FaceDetectorYN wrapper, including an int8-quantized variant.
The pitch is honest engineering rather than leaderboard chasing: detection quality sits below heavy detectors, but the speed and size make it the sensible default for real-time apps, kiosks, and embedded cameras. It detects faces roughly 10×10 to 300×300 pixels, so it covers the typical webcam-and-doorbell range.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: WIDER FACE validation AP 0.834 easy / 0.824 medium / 0.708 hard (OpenCV Zoo's own evaluation reports 0.8498 / 0.8384 / 0.7357). Model is 228KB / 0.083M parameters. No independent published benchmark found.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified, metrics-v1). Evidence: peer-reviewed paper (Machine Intelligence Research, 2023, Springer) behind the model; official weights and evaluation in the OpenCV Zoo; OpenCV's own blog benchmarks it against Haar cascades. Strengths: backed by OpenCV, genuinely tiny and fast, clear honest specs. Docked for: (1) accuracy trails heavyweight detectors and no independent benchmark exists; (2) limited size range for detectable faces. License: model files in the OpenCV Zoo are MIT-licensed.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Peer-reviewed paper (Machine Intelligence Research, Springer 2023); official model in the OpenCV Zoo from Shiqi Yu's libfacedetection.
- · 228KB / 0.083M params; WIDER FACE AP 0.834/0.824/0.708 provider-reported (OpenCV Zoo eval 0.8498/0.8384/0.7357).
- · Faces only ~10x10 to 300x300 px detectable; accuracy below heavyweight detectors.