Detect football players in videos using YOLOv5 for training and YOLOv8 for inference

This project focuses on detecting football players in video footage using state-of-the-art object detection models. The approach leverages YOLOv5 for the training phase and YOLOv8 for inference, combining the strengths of both versions for optimal performance.

The dataset used for this project is sourced from Roboflow and contains:

• 663 annotated images specifically labeled for football player detection
• Format: YOLO v5 PyTorch format
• Pre-processing applied:
  • 50% probability of horizontal flip
  • Random brightness adjustment between -20% and +20%

The dataset is named "football-players-detection" and was exported from Roboflow on December 5, 2022. You can access the original dataset here.

Before running this project, ensure you have the following installed:

• Python 3.x (recommended 3.8 or higher)
• pip (Python package manager)
• Required Python packages:
  • ultralytics
  • opencv-python
  • roboflow

1. Clone the repository:

   git clone https://github.com/RockENZO/ML-object-detection-project.git
   cd ML-object-detection-project

2. Install the required dependencies:

   pip install ultralytics opencv-python roboflow

The training process uses YOLOv5 on the Roboflow dataset. The dataset has already been downloaded and organized in the training/ directory.

To train the model:

# Navigate to the project directory
cd ML-object-detection-project

# Train using the provided data.yaml configuration
yolo task=detect mode=train model=yolov5s.pt data=training/football-players-detection-1/data.yaml epochs=100 imgsz=640

Training parameters can be adjusted as needed:

• epochs: Number of training epochs (default: 100)
• imgsz: Image size for training (default: 640)
• model: YOLOv5 variant to use (yolov5s.pt, yolov5m.pt, yolov5l.pt, etc.)

For inference, the project uses YOLOv8 via the yolo_inference.py script:

# Run the inference script
python yolo_inference.py

This will:

1. Load the YOLOv8s model
2. Process the sample video (input_videos/08fd33_4.mp4)
3. Save the results with bounding boxes to the runs/detect/predict/ directory
4. Print detection results to the console

To run inference on a different video, modify the yolo_inference.py script:

# Change this line to point to your video
results = model.predict('path/to/your/video.mp4', save=True)

ML-object-detection-project/
├── input_videos/                 # Input video files for processing
│   └── 08fd33_4.mp4             # Sample input video
├── models/                      # Pre-trained model weights
│   ├── yolov5su.pt              # YOLOv5 small model
│   └── yolov8s.pt               # YOLOv8 small model
├── runs/                        # Output from training and inference
│   └── detect/                  # Detection results
│       ├── predict/             # Latest inference results
│       ├── predict1/            # Previous inference runs
│       └── predict2/            # Earlier inference runs
├── training/                    # Training dataset and configuration
│   └── football-players-detection-1/
│       ├── data.yaml           # Dataset configuration for YOLO
│       ├── football-players-detection-1/  # Actual dataset
│       │   ├── train/          # Training images and labels
│       │   ├── val/            # Validation images and labels
│       │   └── test/           # Test images and labels
│       ├── README.dataset.txt  # Dataset information
│       └── README.roboflow.txt # Roboflow export details
├── tracker/                     # Object tracking utilities
│   ├── __init__.py
│   └── tracker.py               # Tracking implementation
├── utils/                       # Utility functions
│   ├── __init__.py
│   └── video_utils.py           # Video processing helpers
├── yolo_inference.py            # Main inference script (YOLOv8)
├── main.py                      # Alternative entry point
└── README.md                    # This file

• Training Model: YOLOv5 (selected for its proven performance and stability in training scenarios)
• Inference Model: YOLOv8s (chosen for its improved speed and accuracy in real-time applications)

This dataset is provided by a Roboflow user under the CC BY 4.0 license. Please refer to the original Roboflow dataset page for specific licensing details.

• Roboflow for providing the annotated dataset
• Ultralytics for the YOLOv5 and YOLOv8 implementations
• The open-source computer vision community