Multiple Object Detection using YOLOv8 and Synthetic Data

This repository trains and evaluates a YOLOv8 object detection model using synthetic data and validates it on real-world images. The primary execution path is the Colab notebook syntheticDataWorks_multiclass.ipynb.

What this project does

• Trains YOLOv8 on synthetic training images under Output/train
• Validates using Output/val
• Tests the model on real-world images under testImages
• Saves annotated predictions to Output/predictions
• Generates a summary report using report/report.py

How to run

1. Open syntheticDataWorks_multiclass.ipynb in Google Colab
2. Mount Google Drive when prompted
3. Enable GPU: Edit > Notebook settings > Hardware accelerator > GPU
4. Run all notebook cells from top to bottom

Important notebook behavior

The notebook is designed to:

• mount the signed-in user’s Google Drive
• locate the shared source dataset folder under MyDrive
• copy the source project into the user's own Drive folder at MyDrive/syntheticDataWorks_multiclass
• then proceed with training, prediction, and report generation

Current path setting

If your Drive folder contains the shared project under MyDrive/Multiple_object_detection, the notebook should use:

SOURCE_PROJECT_ROOT = DRIVE_ROOT / 'Multiple_object_detection'

If the source folder path changes, update that line before running.

What each script does

Output/train.py

• Loads the YOLOv8 model from Output/yolov8s.pt
• Uses dataset configuration from Output/yolo_params.yaml
• Trains the model and saves results to Output/runs/detect/train

Output/predict.py

• Loads the best model from the latest training run
• Predicts on images defined by the test field in Output/yolo_params.yaml
• Saves annotated prediction images to Output/predictions/images
• Saves prediction labels to Output/predictions/labels

report/report.py

• Reads training and validation metrics from Output/runs/detect
• Copies example prediction images
• Generates report/generatedreport.md

Outputs

• Output/runs/detect/train — training results and metrics
• Output/predictions/images — annotated real-world predictions
• Output/predictions/labels — predicted label files
• report/generatedreport.md — report summary

Technical Assets

• Runs: Training and validation runs stored in Output/runs/detect/
• Training Logs: Metrics and loss curves in CSV and text formats within run directories
• README Files: This README.md and generated report in report/
• Codes:
  • syntheticDataWorks_multiclass.ipynb — Main Colab notebook
  • Output/train.py — Training script
  • Output/predict.py — Prediction script
  • Output/visualize.py — Visualization utilities
  • report/report.py — Report generation script
• Configuration: Output/yolo_params.yaml — Dataset and model parameters
• Datasets: Synthetic training/validation data and real-world test images with labels

Project structure

├── Output/
│   ├── classes.txt
│   ├── predict.py
│   ├── train.py
│   ├── visualize.py
│   ├── yolo_params.yaml
│   ├── train/
│   ├── val/
│   └── runs/
├── report/
│   └── report.py
├── testImages/
│   ├── images/
│   └── labels/
└── syntheticDataWorks_multiclass.ipynb

Notes

• The notebook assumes the shared folder is accessible from the signed-in Google account.
• If the path is wrong, update SOURCE_PROJECT_ROOT in the notebook.
• The notebook no longer depends on a GitHub clone and uses local Drive storage instead.