https://github.com/headless-start/yolov4-tiny-raspberrypi

This repository contains YOLOv4-Tiny deployment on Raspberry pi.
https://github.com/headless-start/yolov4-tiny-raspberrypi

darknet edge numpy object-detection opencv powerpoint python-3 pytorch raspberry-pi tensorflowlite yolov4 yolov4-darknet yolov4-tiny

Last synced: 1 day ago
JSON representation

This repository contains YOLOv4-Tiny deployment on Raspberry pi.

• Host: GitHub
• URL: https://github.com/headless-start/yolov4-tiny-raspberrypi
• Owner: headless-start
• License: mit
• Created: 2025-01-10T14:41:31.000Z (about 1 month ago)
• Default Branch: main
• Last Pushed: 2025-02-01T08:30:32.000Z (9 days ago)
• Last Synced: 2025-02-08T16:17:41.734Z (1 day ago)
• Topics: darknet, edge, numpy, object-detection, opencv, powerpoint, python-3, pytorch, raspberry-pi, tensorflowlite, yolov4, yolov4-darknet, yolov4-tiny
• Homepage:
• Size: 3.75 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Real-Time Object Detection on Edge Devices with YOLOv4-Tiny  

## 📌 Project Overview  

This project analyzes the suitability of **YOLOv4-Tiny** for real-time object detection on **edge devices** like the Raspberry Pi. The lightweight architecture of YOLOv4-Tiny enables high-speed inference, making it ideal for resource-constrained hardware. This work is presented in the form of a **college task**, focusing on the architectural advantages and performance optimizations of YOLOv4-Tiny over vanilla YOLOv4.  

---

## 🚀 Key Features  

1. **Lightweight Architecture**:  

   - YOLOv4-Tiny uses fewer layers and smaller backbones, making it faster and more efficient for edge deployment.  

2. **Real-Time Performance**:  

   - Optimized for speed and efficiency while maintaining acceptable accuracy for real-world applications.  

3. **Edge Device Optimization**:  

   - Tested on devices like Raspberry Pi 4B for optimal detection.

---

## 🛠 Architecture and Performance Enhancements  

### YOLOv4-Tiny Architecture  

1. **Backbone**:  

   - CSPDarknet53-Tiny for efficient feature extraction.  

2. **Neck**:  

   - Feature Pyramid Network (FPN) for feature pooling.  

3. **Head**:  

   - Dual-scale detection for large and small objects using 13×13 and 26×26 grids.  

### Performance Enhancements  

1. **Fewer Parameters**:

   - ~6 million compared to ~63.6 million in YOLOv4.  

3. **Simpler Activation**:

   - Uses LeakyReLU instead of Mish activation for faster computation.  

5. **Quantization and Pruning**:

   - Reduces model size and increases speed for edge deployment.  

7. **TPU Acceleration**:

   - Compatible with hardware like Google Coral for faster inference.  

---

## 🛠 Usage  

### Dependencies  

- **Python**: 3.8+  

- **Libraries**: `tensorflow-lite` ,`darknet` , `numpy`, `opencv-python`

- **Hardware**: Raspberry Pi 4B 

---

## 📄 License  

This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.