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

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 (YOLOv8)

Streamlit Demo Sample