Home

📌 Features

• 🧠 UNet (Encoder-Decoder) Architecture
• 🔁 Custom Dataset Loader with Label Remapping
• 📊 Evaluation using IoU, Pixel Accuracy
• ⚡ Optimized Training Pipeline (CPU/GPU support)
• 🖼️ Before vs After Segmentation Visualization

---

🧠 Model Architecture

• Encoder → Feature Extraction
• Bottleneck → Deep Representation
• Decoder → Spatial Reconstruction
• Skip Connections → Preserve details

---

📂 Project Structure

offroad-segmentation/ │ ├── data/ │ ├── train/ │ ├── val/ │ └── test/ │ ├── models/ │ └── unet.py │ ├── utils/ │ ├── dataset.py │ └── metrics.py │ ├── train.py ├── evaluate.py ├── predict.py ├── config.py ├── requirements.txt └── README.md

---

⚙️ Setup

1. Clone the repository

git clone https://github.com/your-username/offroad-segmentation.git
cd offroad-segmentation
2. Install dependencies
pip install -r requirements.txt
▶️ Usage
🔹 Train Model
python train.py
🔹 Evaluate Model
python evaluate.py
🔹 Run Inference (Demo)
python predict.py
📊 Results
Metric	Value
Mean IoU	~0.42
Pixel Accuracy	~0.81
Approx mAP	~0.79

Note: mAP is approximated for segmentation (not standard detection mAP)

🖼️ Sample Output

Input Image → Segmentation Output

(Add your result.png here)

⚠️ Important Notes
Dataset is not included due to size constraints
Model is trained only on provided dataset (as per hackathon rules)
No external data used
🧠 Key Learnings
Handling non-contiguous class labels
Building custom dataset pipelines
Optimizing training on limited hardware
Evaluating segmentation models effectively
🚀 Future Improvements
Data Augmentation
Advanced Models (DeepLabV3+)
Better class balancing
Real-time inference
👨‍💻 Team
[Nikhil kandhare(Team GCOEY)]
⭐ Acknowledgements
Duality AI Hackathon Dataset
PyTorch Community
📬 Contact
nikhilkandhare22@gmail.com
+91 9112430021