YOLOv5 Medium, YOLOv8 Nano Model Trained on 5 Classes

This repository contains a YOLOv5, YOLOv8n model trained on a dataset that includes 5 classes: Person, Bus, Car, Motorbike, and Bicycle. YOLO (You Only Look Once) is a popular object detection model capable of real-time object detection. The "Medium" variant of YOLOv5 refers to the specific architecture and model size used in this implementation.

Input: image or camera

Ouput: processed image or frame

Table of Contents

• YOLOv5 Medium, YOLOv8 Nano Model Trained on 5 Classes
• Table of Contents
  • Model Details
  • Installation and Dependencies
  • Usage
    • YoloV5
    • YoloV8
  • Model Performance
  • Training
  • Acknowledgments
  • Author
  • Contact

Model Details

• Model Size: Medium
• Classes: 5 (Person, Bus, Car, Motorbike, Bicycle)
• Framework: PyTorch
• Input Image Size: 448x448
• Hyperparameters: lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=0.05, cls=0.5, cls_pw=1.0, obj=1.0, obj_pw=1.0, iou_t=0.2, anchor_t=4.0, fl_gamma=0.0, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, mosaic=1.0, mixup=0.0, copy_paste=0.0

Installation and Dependencies

pip install requirements.txt

Usage

To use this YOLOv5 model for transportation detection, follow these steps:

1. Clone this repository:

git clone https://github.com/hotonbao/YOLOV5.git
cd YOLOV5

2. Dowload Weight File:

   Weight YoloV5: https://drive.google.com/file/d/1zw0rR7iSfobJ9CwPXe2-YqvjrSmjzt_T/view?usp=sharing

   Weight YoloV8: https://drive.google.com/file/d/1OSU5g3yz-IliqMFQhfbmVp0UA6e5nKcT/view?usp=sharing

   Notice: Replace PATH_MODEL in the configs.py(for YoloV5) file with the pre-trained model file.

3. Run the inference script to perform object detection on an image with FastAPI or Streamlit(YoloV5):

   YoloV5

   FastAPI:

   python api.py

   Type "/docs" after link http://127.0.0.1:8000 to test with any image

   Streamlit:

   Notice: Run FastAPI before run Streamlit and open new command line and run:

   streamlit run app.py --server.fileWatcherType=none

   It will open:

   

   YoloV8

   cd yolo-v8
   python app.py

   It will open a window:

   

   It have 3 functions:

   • Load Image
   • Start Camera
   • Stop Camera

4. -Ouput shape after inference:

   

   • 1: is number of batch

   • 12348: number of bounding boxes in image.

   • 10=(4+1+5):

     • x_max, y_max, x_min, y_min (4)
     • confidence score is an object (1)
     • confidence correspond to each class(5).

   -Output postprocess Run test.ipynb file to see output.

   Example:

   

   It is an object that has 6 arguments equivalent to

   [ xmin, ymin, xmax, ymax, confidence of class, classname ']

   Description: Confidence score: Prediction score of class.

   Class name: has been encoded : 0: "bicycle", 1: "bus", 2: "car", 3: "motorbike", 4: "person"

Model Performance

• Loss
• F1 Score For Class :
• 

Training

Dataset: https://www.kaggle.com/datasets/yusufberksardoan/traffic-detection-project/

Load the dataset link in and enter yolov5-training.ipynb run all cells then download the best m.onnx file.

Acknowledgments

https://github.com/ultralytics/yolov5/tree/master/models

Visit the link if you want to train the model with many different architectural sizes

Author

Ho Ton Bao

Contact

Phone number : +84949800149

Email: [email protected]

---