skills.rest
limit5
limit5COMMUNITY
Explore all skills by this author →
GitHub

npu-recognition

Deploy face recognition models to NPU hardware with INT8 quantization and FAR/FRR validation.

1|Updated Apr 9, 2026
Search Tags:#quantization#embedding#identity-verification#face-recognition#npu#arcface#far-frr
One-click install
npx skills add https://github.com/limit5/OmniSight-Productizer --skill npu-recognition
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: npu-recognition
Source: https://github.com/limit5/OmniSight-Productizer/tree/main/configs/skills/npu-recognition
Command: npx skills add https://github.com/limit5/OmniSight-Productizer --skill npu-recognition

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Deploying face and identity recognition models on NPU hardware often faces challenges with quantization accuracy loss, latency failures, and integration complexity, making it difficult to meet production-grade performance and reliability requirements.

Core Features & Use Cases

  • End-to-End NPU Deployment Workflow: Covers pipeline setup, model quantization, accuracy verification, and system integration for face recognition use cases.
  • Production-Grade Validation: Built-in FAR/FRR metric checks and embedding drift thresholds to ensure models meet strict accuracy and latency requirements before deployment.
  • Use Case: Deploy a face access control system for an embedded smart camera that verifies user identities in under 50ms with a false accept rate below 0.001.

Quick Start

Use the npu-recognition skill to deploy an ArcFace face recognition model to NPU hardware with INT8 quantization and verify it meets the 0.001 FAR threshold on the LFW dataset.

Frequently Asked Questions about npu-recognition

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I deploy an ArcFace face recognition model to NPU hardware with INT8 quantization?▼

To deploy an ArcFace face recognition model to NPU hardware, use an end-to-end workflow that handles ONNX model compatibility, applies INT8 quantization, and verifies accuracy metrics like FAR before finalizing the embedded vision deployment.

What is the best way to validate face recognition accuracy and false accept rate on NPU before production deployment?▼

Validating face recognition accuracy on NPU requires running built-in FAR and FRR metric checks alongside embedding drift thresholds, ensuring the model meets strict latency and reliability requirements before production deployment.

Can I use ONNX models for 1:1 identity verification and 1:N identification on embedded smart cameras?▼

Yes, ONNX models can be deployed for 1:1 identity verification and 1:N identification on embedded smart cameras, utilizing NPU hardware for low-latency feature extraction and cosine similarity matching with an embedding database.

Why does face recognition quantization cause accuracy loss on NPU, and how can I verify FP16 or INT8 performance?▼

Face recognition quantization can cause accuracy loss on NPU due to reduced precision, but you can verify FP16 or INT8 performance by checking FAR and FRR metrics against validation datasets like LFW to ensure accuracy thresholds are met.

Does this NPU face recognition workflow support embedding database integration for access control systems?▼

Yes, the NPU face recognition workflow supports embedding database integration, enabling smart camera access control and identity verification systems to perform low-latency cosine similarity matching for user identification.

Related Skills

Explore More Software Engineering Skills→
igamenovoerigamenovoer
7

pixi-make-cu-build-env

Configure CUDA build environments for Pixi projects with CMake and nvcc.

Community
Advanced
discountedcookiediscountedcookie

postgres-vectors

Store 384-dimensional embeddings in Postgres and query cosine similarity with pgvector.

Community
Advanced
autonomous-aiautonomous-ai
10

face-enroll

Manage the lamp's face recognition roster via HTTP endpoints.

Official
Advanced
amdamd
154

quark-onnx-quant-plan

Build hardware-aware Quark ONNX quantization plans from model analysis and deployment intent.

Official
Advanced