YOLOv8-based object detection model for identifying doors and windows in architectural blueprints. Developed as part of an assignment for Palcode.ai.

---

🔧 Step 1: Manual Labeling

• Dataset manually labeled using Roboflow.
• Classes: door, window.

---

📁 Step 2: Dataset Preparation & Augmentation

• Total images: 20
• Train/Validation Split:
  • Training set: 16 images
  • Validation set: 4 images

✅ Preprocessing

• Auto-orient: ✅
• Resize: Stretched to 1024×1024

🔁 Augmentations

• Outputs per training image: 3
• 90° Rotate: Clockwise, Counter-clockwise, Upside-down
• Random Rotation: Between -15° and +15°
• Brightness Adjustment: ±15%
• Exposure Adjustment: ±10%
• Noise: Up to 0.1% of pixels

📦 Exports

Two dataset versions were exported for evaluation:

• 640×640 resolution
• 1024×1024 resolution

---

🤖 Step 3: Model Training & Evaluation

Three models were trained and evaluated for performance:

Model	Notes
YOLOv8n	Performed best overall
YOLOv8s	Lower accuracy
YOLOv11n	Lower accuracy

After repeated fine-tuning, YOLOv8n delivered the best results in terms of precision and bounding box accuracy.

---

📝 Summary

• Labeled with Roboflow
• Augmented with robust transformations
• Trained on YOLOv8n
• Ready for deployment as an ONNX model

---

Run Flask API:

pip install flask
python app.py