SCRFD
The InsightFace team's answer to efficient face detection: redistribute samples and compute instead of scaling the model.
About
SCRFD (Sample and Computation Redistribution for Face Detection) argues that most efficient detectors waste compute in the wrong places. It contributes two ideas: sample redistribution, which augments training data toward the stages that need it most, and computation redistribution, which reallocates FLOPs across backbone, neck, and head using a searched design space.
The result is a whole family of detectors covering compute regimes from 0.5 to 34 GFLOPs, all with state-of-the-art efficiency-accuracy trade-offs on WIDER FACE. SCRFD powers the detection stage of InsightFace's buffalo model packs, which is why it quietly runs in thousands of production attendance and kiosk systems.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: SCRFD-34GF beats the prior best competitor TinaFace by 3.86–4.78% AP on the WIDER FACE hard set while running 3x faster on GPU at VGA resolution. 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 ICLR 2022 paper (arXiv 2105.04714) by the InsightFace team; official code and weights in the InsightFace repo; powers the widely deployed buffalo_l/buffalo_s model packs. Strengths: top-tier venue, broad real-world deployment through InsightFace, clear efficiency story. Docked for: (1) no independent published benchmark; (2) effectively a component of InsightFace rather than a standalone supported product. License: InsightFace code is MIT; weight licensing is research-oriented.
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
- · ICLR 2022 (arXiv 2105.04714) by Guo, Deng et al. (InsightFace / Imperial College).
- · SCRFD-34GF beats TinaFace by ~3.9-4.8% AP on WIDER FACE hard at 3x GPU speed, provider-reported.
- · Detection backbone of InsightFace buffalo model packs; no independent benchmark found.