Traffic Sign Detection: YOLOv8n on TT100K

Compact traffic-sign detection project for an Automated Driving System capstone.
The model is trained on TT100K (50 classes) with YOLOv8n and exported to ONNX.

Features

• YOLOv8n training pipeline for TT100K (50 classes)
• Config-driven workflow via YAML
• Validation and single-image inference commands
• ONNX export script for Raspberry Pi CPU deployment
• Kaggle notebook + modular Python pipeline
• CI + tests scaffolding

Project Structure

traffic-sign-detection-model/
├── .github/workflows/ci.yml
├── configs/
│   ├── default.yaml
│   └── kaggle.yaml
├── data/
├── notebooks/
├── scripts/
│   ├── export_onnx.py
│   └── infer_image.py
├── src/
│   ├── __init__.py
│   ├── pipeline/
│   │   ├── __init__.py
│   │   ├── evaluate.py
│   │   └── train.py
│   └── utils/
│       ├── __init__.py
│       └── io.py
├── tests/
│   └── test_config.py
├── .gitignore
├── LICENSE
├── requirements.txt
└── run.py

Dataset

For local training, place the dataset under:

• data/tt100k/mydata

Expected layout:

data/tt100k/mydata/
├── images/
│   ├── train/
│   └── val/
└── labels/
    ├── train/
    └── val/

If you want to download the Kaggle version manually, use the Kaggle API and unzip it into data/tt100k/mydata. The Kaggle notebook can still use the Kaggle-hosted path directly through configs/kaggle.yaml.

Kaggle path used by the notebook:

/kaggle/input/datasets/braunge/tt100k/mydata

Quick Start

1. Install dependencies:

pip install -r requirements.txt

2. Download or copy the dataset so it exists at data/tt100k/mydata.

3. Train locally:

python run.py --config configs/default.yaml --mode train

4. Validate:

python run.py --config configs/default.yaml --mode validate

5. Export ONNX:

python run.py --config configs/default.yaml --mode export

6. Inference on one image:

python run.py --config configs/default.yaml --mode infer --image /path/to/image.jpg --conf 0.25

Notes

• The default config auto-selects the device when possible, so it works on GPU machines and CPU-only machines.
• If you are on Kaggle, use configs/kaggle.yaml instead of the local default.
• Training resumes from the latest checkpoint in runs/detect/tt100k_training/yolov8n_30e/weights/last.pt when that file exists.

License

Apache-2.0. See LICENSE.