🚀 Awesome NCNN Collection

⚡ High-Performance Neural Network Inference for Mobile & Edge AI

Your ultimate curated collection of cutting-edge NCNN resources, projects, and implementations

🌟 Star • 🔥 Trending Projects • 📱 Mobile Deploy • ⚙️ Optimization

---

📊 Performance Highlights

Metric	Performance	vs GPU
⚡ Latency	58.54% Lower	Matrix Operations
🚄 LLM Inference	3.2× Faster	Inference Speed
💪 Throughput	2× Higher	vs NPUs
📱 Platforms	10+ Supported	Cross-Platform

---

📑 Table of Contents

🗂️ Click to expand/collapse

• 🎯 Official NCNN Resources
• 🔥 Latest YOLO Implementations (2024-2025)
• 🎯 Object Detection
• 🖼️ Image Classification
• ✂️ Image Segmentation
• ✨ Super Resolution & Image Enhancement
• 👤 Face & Biometrics
• 🏃 Pose Estimation & Human Tracking
• 🎙️ Speech Recognition & ASR
• 📝 OCR & Text Recognition
• 🎬 Video Processing
• 🎨 Stable Diffusion & Generative Models
• 📦 Model Collections & Zoo
• 📱 Platform-Specific Deployment
• ⚙️ Optimization & Quantization
• 🛠️ Tools & Utilities
• 📚 Tutorials & Documentation
• 📊 Benchmarks & Performance
• 📄 Research Papers (2024-2025)

---

🎯 Official NCNN Resources

🏆 Core Framework

🔧 Main Repository ⭐ Tencent/ncnn 22k+ stars - Official framework Mobile-optimized inference

📖 Documentation 📘 Official Docs 📙 DocsForge 🌋 Vulkan FAQ 📊 Benchmarks

🐍 Python Integration

• 🔗 NCNN Python Wrapper - Official Python bindings

---

🔥 Latest YOLO Implementations (2024-2025)

🏅 YOLOv11 (Latest - 2024)

📱 Production-Ready Implementations

Project	Platform	Performance	Description
📘 Ultralytics Official	All	⚡ Optimized	Official NCNN export guide
⚙️ yolo11-ncnn	C++	48ms	High-performance C++ impl
🎯 No Magic Ops	All	✅ Clean	Direct Ultralytics export
🤖 Android YOLOv11	Android	📱 Mobile	Optimized for Android
🎥 Real-time Demo	Android	🚀 Fast	Live detection demo
🤖 ROS 2 Support	ROS	🔧 Robotics	YOLOv8/v9/v10/v11

🎖️ YOLOv10 & YOLOv9 (2024)

• 📊 Performance Comparison - v8/v9/v10/v11 benchmarks

🥈 YOLOv8 & Earlier Versions

📦 Click to expand YOLOv5-v8 implementations

• 🎯 YOLOv8 Android
• 🎯 YOLOv6 Android
• ✂️ YOLOv5 Segmentation
• 📱 YOLOv6 + Guide
• 🍎 YOLOv5 iOS/Android
• ⚡ YOLOv8 TensorRT

🌟 Other YOLO Variants

• 🏢 YOLOv6 Official - Meituan's production YOLO
• 🏭 DAMO-YOLO - Industrial applications
• 🌐 YOLO-Universal

---

🎯 Object Detection

🌟 NanoDet (Lightweight Detection)

⚡ Super fast and ultra-lightweight anchor-free detection

• 🚀 NanoDet -
• 📱 Android Demo - Official NCNN implementation

👤 BlazeFace

• 🔥 BlazeFace Android - Google's face detection

🎁 General Detection

• 🎁 Lite.AI.ToolKit - Comprehensive AI toolkit
• 📱 YOLO Mobile - Mobile-optimized YOLO

---

🖼️ Image Classification

🧠 Classic Models 🔬 SqueezeNet 📱 MobileNet 🚀 MobileNetV2

📊 Features ✅ Lightweight architectures ⚡ Fast inference 📱 Mobile-optimized

---

✂️ Image Segmentation

🎭 Portrait & Semantic Segmentation

Model	Use Case	Platform	Stars
👤 Portrait Seg	Portrait	Mobile
🎬 Video Matting	Real-time	Android
🎯 SAM Android	Segment Anything	Android	🆕 2024
🏥 U-Net	Medical/General	All	Classic
🎨 BiSeNet	Semantic	All	SOTA

---

✨ Super Resolution & Image Enhancement

🌟 Real-ESRGAN & Waifu2x Family

🎨 Production-Ready Solutions

🖼️ Real-ESRGAN Series 🎯 Real-ESRGAN Vulkan GPU accelerated 4K+ upscaling Production-ready 📱 Android All-in-One RealSR, SRMD, RealCUGAN Real-ESRGAN, Waifu2x Anime4k support

🎌 Waifu2x & Anime 💎 Waifu2x GUI Video/Image upscaling Frame interpolation Batch processing 🎨 Original Waifu2x Anime-style art Noise reduction Classic model

🔬 Advanced Models

• 🌟 SRMD Vulkan - Super-resolution
• 📸 RealSR - Real-world SR
• 🎨 Picture Enhancement - General enhancement
• 👤 GFPGAN - Face restoration

---

👤 Face & Biometrics

🔍 Face Detection & Recognition

Model	Size	Speed	Accuracy
🏆 Ultra-Light	1MB	⚡⚡⚡	⭐⭐⭐⭐
🎯 PFLD	2MB	⚡⚡	⭐⭐⭐⭐⭐

• 🥇 Ultra-Light-Fast 1MB

  • 
  • Ultra lightweight
  • Real-time detection

• 📍 PFLD Landmark

  • 98 facial landmarks
  • High accuracy
  • Mobile-optimized

👁️ Other Biometrics

• 🎯 Iris Landmarks - Eye tracking

---

🏃 Pose Estimation & Human Tracking

🎯 State-of-the-Art Models

🌟 Featured Models

🔥 RTMPose Snapdragon 865 NCNN optimized High accuracy

🎯 YOLOv8-Pose Real-time Multi-person Edge devices

⚡ BlazePose MediaPipe based 33 keypoints Mobile-first

📚 Additional Models

• 🏃 MoveNet - Efficient single/multi-pose
• 🎭 PoseNet - Classic pose estimation
• 📱 MobileNet Pose - Lightweight variant

📊 Resources

• 📈 Best Models 2024 - Comprehensive comparison

---

🎙️ Speech Recognition & ASR

🎤 Sherpa-NCNN - Next-Gen Speech Recognition

🚀 Main Repository

• 🎯 k2-fsa/sherpa-ncnn

✨ Key Features

📱 Platform Support ✅ iOS ✅ Android ✅ Linux ✅ macOS ✅ Windows ✅ Raspberry Pi ✅ VisionFive2 ✅ LicheePi4A

🌍 Language Support 🇨🇳 Chinese 🇺🇸 English 🇯🇵 Japanese 🇰🇷 Korean 🇭🇰 Cantonese

🎯 Features

• ⚡ Real-time recognition
• 🔇 Voice Activity Detection (VAD)
• 🚫 No internet required
• 🎓 Next-gen Kaldi based

📖 Documentation

• 📚 Complete Docs - Pre-trained models & guides

---

📝 OCR & Text Recognition

📖 Lightweight OCR Solutions

🏆 Featured Projects

Project	Languages	Platform	Highlights
🇨🇳 chineseocr_lite	Chinese	All	Super lightweight, vertical text
🔧 ncnn_ocr	Multi	Mobile	Profiling tool
🌐 WebAssembly OCR	Multi	Browser	Browser-based

📊 Resources

• 📈 Top 10 OCR Models 2025

---

🎬 Video Processing

🎥 Frame Interpolation & Optical Flow

🌟 RIFE - Real-Time Frame Interpolation

• ⚡ RIFE-ncnn-vulkan - Production-ready Vulkan implementation
• 📊 ECCV 2022 RIFE
  • 
  • Real-time interpolation
  • 60+ FPS output
  • Anime & video support

✨ Features

• 🎯 Optical flow estimation
• 🎬 Video frame interpolation
• ⚡ GPU accelerated
• 📱 Mobile deployment

---

🎨 Stable Diffusion & Generative Models

🖼️ AI Art Generation on Mobile

🎯 Production Implementations

🎨 Stable Diffusion NCNN EdVince/SD-NCNN ✅ txt2img support ✅ img2img support ✅ C++ implementation 📱 Mobile-ready

🖌️ Features 🎯 512x512 generation ⚡ Optimized inference 💾 Low memory usage 🔥 Vulkan acceleration

📚 Reference

• 🌐 SD WebUI - Desktop reference

---

📦 Model Collections & Zoo

🗃️ Pre-trained Model Libraries

🏆 Curated Collections

Repository	Models	Guides	Stars
🎁 ncnn-models	100+	✅
📦 ncnn_models	50+	✅
🏛️ ncnn-assets	Official	✅	Official
⭐ awesome-ncnn	Curated	✅

🌐 Platform Support

• 💻 Windows, Mac, Linux
• 📱 Android, iOS
• 🌐 WebAssembly
• 📲 Uni-app

---

📱 Platform-Specific Deployment

🤖 Android

📱 Click to expand Android resources

📚 Comprehensive Guides

• 📖 PyTorch → NCNN Android

  • Complete deployment tutorial
  • NDK setup guide
  • CMake configuration

• 📦 ncnn-mobile

  • Android + iOS support
  • Sample projects
  • Best practices

🔧 Requirements

• ✅ Android NDK
• ✅ CMake 3.10+
• ✅ OpenCV Android (optional)

🍎 iOS

📱 Click to expand iOS resources

📚 Swift Integration

• 🔗 C++ in Swift Guide

  • Bridging header setup
  • Framework integration
  • Best practices

• 🎁 ncnn-swift

  • Swift wrapper
  • CocoaPods support
  • Sample apps

⚙️ Configuration

• ✅ Xcode 12+
• ✅ iOS 11+
• ✅ Vulkan/Metal support

🌐 WebAssembly

🌐 Click to expand WASM resources

• 🎯 Portrait Segmentation WASM
  • Browser-based inference
  • No server required
  • Real-time processing

🚀 Features

• ✅ Zero installation
• ✅ Privacy-preserving
• ✅ Cross-browser support

🔧 OpenCV Integration

• 📷 opencv-mobile
  • Minimal OpenCV build
  • NCNN optimized
  • Reduced size

---

⚙️ Optimization & Quantization

🚀 Performance Optimization

🎯 INT8 Quantization

📚 Official Resources

• 📖 INT8 Inference Guide
  • Post-training quantization
  • Calibration workflow
  • KL & ACIQ methods

🛠️ Tools & Libraries

🔧 Conversion Tools ⚡ caffe-int8-tools 🎓 ncnnqat - QAT support

📊 Benefits ⚡ 4× faster inference 💾 4× smaller models 📱 Lower memory usage 🔋 Better battery life

🌋 Vulkan Optimization

📚 Resources

• 🔥 NCNN Machine Vision 2025
• ⚙️ Vulkan Optimization Notes

✨ Key Features

• ✅ GPU acceleration
• ✅ Cross-platform
• ✅ Memory optimization
• ✅ Pipeline caching

---

🛠️ Tools & Utilities

🔄 Model Conversion & Editing

🔧 Conversion Tools ⚡ ONNX Simplifier Model optimization Graph simplification Size reduction 🌐 ConvertModel.com Online converter Multiple formats Free to use

🎨 Visual Tools 🖼️ ncnn-editor Visual model editing Graph visualization Debug support 📊 ncnn-small-board Model visualization Layer inspection Performance profiling

🔗 Inference Helpers

• 🎁 InferenceHelper

  • Unified interface
  • Multiple frameworks
  • Easy integration

• 📦 InferenceHelper_Sample

  • Sample projects
  • Best practices
  • Quick start

🏗️ Build Tools

• 🔨 CMake Examples
  • CMake integration
  • Cross-platform builds
  • Modern CMake

---

📚 Tutorials & Documentation

📖 Official Documentation

🇺🇸 English Resources 📘 NCNN Wiki 🔨 How to Build 📖 PyTorch → NCNN Part 1 📱 PyTorch → NCNN Part 2 🆚 PyTorch vs ONNX vs NCNN 🚀 Getting Started

🇨🇳 Chinese Resources 🔬 ncnn_breakdown 📝 Zhihu Column 📚 CSDN Analysis 📖 Deep Dive Series

---

📊 Benchmarks & Performance

⚡ Official Benchmark Results (Dec 2024)

🏆 Key Metrics

Metric	NCNN	Comparison	Improvement
⚡ Latency	Low	vs GPU	58.54% ↓
🚄 LLM Inference	Fast	vs Baseline	3.2× ↑
💪 Throughput	High	vs NPU	2× ↑
📱 Platforms	10+ Supported

📊 Benchmark Resources

• 📈 Official Benchmarks
• 🔥 OpenBenchmarking

🆚 Framework Comparisons

🔬 embedded-ai.bench Benchmark Suite NCNN vs TNN vs MNN TFLite comparison

📱 Mobile AI Bench Xiaomi Benchmark Real device testing Production workloads

⚡ vs TensorRT Detailed Comparison Use case analysis Performance metrics

---

📄 Research Papers & Articles (2024-2025)

📚 Latest Academic Research

🎯 YOLO Family Comparisons

📊 Performance Studies 📈 YOLOv8/v9/v10/v11 Comparison Comprehensive evaluation Multiple datasets Production insights 🏗️ YOLOv11 Architecture Key enhancements Design principles Technical deep dive

🔥 Application Papers 🚨 Smoke/Fire Detection YOLOv9/v10/v11 Real-world testing Safety applications 🌱 Weed Detection - Agricultural AI 📄 Document Analysis - OCR pipelines 🖼️ Stamp Detection - Document processing

🚀 Mobile AI Trends

• 🎯 NCNN Machine Vision 2025

  • Latest optimization techniques
  • Industry trends
  • Best practices

• 🧠 Post-Transformer Architectures

  • State Space Models
  • RWKV architectures
  • Future directions

---

🤝 Contributing

We welcome contributions! 🎉

How to contribute:

• ⭐ Star this repository
• 🍴 Fork and submit PRs
• 🐛 Report issues
• 💡 Suggest new resources

---

📜 License

This collection is provided as-is for educational and reference purposes.

---

⏰ Last Updated

🗓️ January 2025 - Continuously updated with the latest NCNN projects and resources

🌟 Star History

---

Made with ❤️ for the NCNN Community

⬆ Back to Top