Skip to content

This project integrates MobileViTv3 into YOLOv8 for UAV-based object detection, achieving higher accuracy and smoother training than the original yolov8n.pt. Designed with lightweight efficiency, it is well-suited for deployment on edge devices such as drones. A Streamlit app is provided for intuitive model visualization and real-time inference.

License

Notifications You must be signed in to change notification settings

JSLEE-0703/Yolov8-drone-detection

Repository files navigation

YOLOv8 + MobileViTv3

Author

YuTong Li (Harry)

Requirements

  • 12GB NVIDIA GPU
  • Python 3.9.0
  • PyTorch v1.12 or later

Note: app.py is the Streamlit launcher.
Before running, please make sure you have saved the trained model in the current directory.

About MobileViTv3

MobileViTv3 is introduced in the paper:
MobileViTv3: Mobile-Friendly Vision Transformer with Simple and Effective Fusion of Local, Global and Input Features
by Shakti N. Wadekar and Abhishek Chaurasia

Backbone

Backbone

Contribution of This Project

This project verifies the effectiveness of incorporating MobileViTv3 into YOLOv8:

  • Results are better than the original yolov8n.pt
  • Training curves are smoother

Results

Training and Validation Loss Curves (YOLOv8 + MobileViT)

Training and Validation Loss Curves For YOLOv8+MobileVIT

Training and Validation Loss Curves (YOLOv8)

Training and Validation Loss Curves For YOLOv8

Streamlit Demo Sample

Streamlit Demo Sample

About

This project integrates MobileViTv3 into YOLOv8 for UAV-based object detection, achieving higher accuracy and smoother training than the original yolov8n.pt. Designed with lightweight efficiency, it is well-suited for deployment on edge devices such as drones. A Streamlit app is provided for intuitive model visualization and real-time inference.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published