---
title: LittleHornet-Video-Face-ONNX
canonical_url: "https://www.modelscope.cn/models/BeePlug/LittleHornet-Video-Face-ONNX"
md_url: "https://www.modelscope.cn/models/BeePlug/LittleHornet-Video-Face-ONNX.md"
repository: BeePlug/LittleHornet-Video-Face-ONNX
chinese_name: "小黄蜂视频人物检索 ONNX 模型"
last_updated: 2026-09-13
pipeline_tag: face-recognition
tasks:
  - face-recognition
library_name:
  - onnx
downloads: 0
stars: 0
---

# LittleHornet-Video-Face-ONNX

> LittleHornet-Video-Face-ONNX - BeePlug 在 ModelScope 开源的模型。VideoSceneMaster 人物搜索的按需模型组件，包含 SCRFD-10G 人脸检测与 ArcFace R50 人脸特征模型。模型运行环境另行提供。

- **Repository**: BeePlug/LittleHornet-Video-Face-ONNX
- **Tasks**: face-recognition
- **Downloads**: 0
- **Stars**: 0
- **Last updated**: 2026-09-13

Source: https://www.modelscope.cn/models/BeePlug/LittleHornet-Video-Face-ONNX

---

# LittleHornet Video Face ONNX

VideoSceneMaster optional face detection and face embedding model assets.
This package does not contain video files, portraits, biometric templates,
runtime DLLs or a standalone application. Download only when using face search.

## Files

- `det_10g.onnx`: SCRFD-10G face detector with five landmarks.
- `w600k_r50.onnx`: ArcFace R50, aligned 112x112 RGB input, L2-normalized 512D output.
- `config.json`: model contract, not runtime tuning or a performance guarantee.
- `model-manifest.json`: per-file sizes and SHA-256 hashes.
- `MODEL-PROVENANCE.md`: original local model provenance and usage notices.

Source package: InsightFace buffalo_l.
https://github.com/deepinsight/insightface/releases/tag/v0.7

Original pretrained-weight restrictions remain applicable. This staging package
does not grant or assert commercial or redistribution authorization. The
publisher is responsible for confirming applicable permissions before upload.
No accuracy, speed, DirectML compatibility or production readiness is asserted.

The application will store extracted face data locally and associate observations
with source video times and scene intervals. Model changes require rebuilding
face features; these vectors are not interchangeable with ResNet/SSCD/CLIP.