---
license: mit
language:
- en
pipeline_tag: keypoint-detection
---
# 🅿 Parkospace — AI Parking Space Detector

A lightweight **MobileNetV2-based keypoint detector** that automatically detects the 4 corners of a parking space rectangle in a photo, estimates real-world dimensions, and calculates rental pricing. Built for the [Parkospace](https://github.com/your-username/parkospace) platform.

---

## 🧠 What It Does

- 📸 Takes a photo of a parking space
- 🔍 Detects the 4 corner keypoints of the space (TL, TR, BR, BL)
- 📐 Estimates real-world dimensions in metres (using 2-photo perspective method)
- 💰 Calculates Hourly / Daily / Monthly pricing automatically
- ✏️ Lets users interactively adjust the rectangle (resize + move)

---

## 🏗️ Model Architecture

| Property | Value |
|---|---|
| Architecture | MobileNetV2 + Regression Head |
| Input | `(B, 3, 224, 224)` — normalized RGB |
| Output | `(B, 8)` — 4 corners `(x, y)` normalized to `[0, 1]` |
| Corner order | TL → TR → BR → BL |
| Loss function | Wing Loss |
| Total parameters | ~3.4M |
| Model size | ~13 MB |
| Inference speed | < 30ms on CPU |

---

## 📁 File Structure

```
parkospace/
├── parkospace_colab.ipynb   # Complete Colab notebook (train + test + infer)
├── generate_data.py         # Synthetic parking image generator
├── model.py                 # CNN model definition
├── train.py                 # Training script
├── infer.py                 # Inference + interactive editor + pricing
├── upload_to_hf.py          # HuggingFace Hub upload script
├── requirements.txt         # Dependencies
└── README.md
```

---

## 🚀 Quickstart

### Option 1 — Google Colab (Recommended)
Open `parkospace_colab.ipynb` in [Google Colab](https://colab.research.google.com), switch to **T4 GPU**, and run all cells top to bottom. No setup needed.

### Option 2 — Run Locally

**1. Clone and install**
```bash
git clone https://github.com/your-username/parkospace-detector
cd parkospace-detector
pip install -r requirements.txt
```

**2. Generate synthetic training data**
```bash
python generate_data.py --out data --n 5000
```

**3. Train the model**
```bash
python train.py --data data --epochs 30 --batch 32
```

**4. Run inference on an image**
```bash
python infer.py --checkpoint checkpoints/best.pt --img parking.jpg
```

**5. Dual-image dimension estimation**
```bash
python infer.py \
  --checkpoint checkpoints/best.pt \
  --length_img  photo_from_length_side.jpg \
  --breadth_img photo_from_breadth_side.jpg
```

---

## 🐍 Python API

```python
from model import ParkingSpaceDetector
from infer  import load_model, detect, calculate_pricing

# Load trained model
model = load_model("checkpoints/best.pt")

# Detect parking space corners
import cv2
img = cv2.imread("parking.jpg")
det = detect(model, img)

print("Corners (pixels):", det.corners_px)
# [[x0,y0], [x1,y1], [x2,y2], [x3,y3]]  →  TL, TR, BR, BL

# Calculate pricing
pricing = calculate_pricing(area_m2=15.0)
print(pricing)
# {
#   "hourly_inr":  50.0,
#   "daily_inr":   300.0,
#   "monthly_inr": 150.0,
#   "area_m2":     15.0
# }
```

---

## 📐 Dual-Image Dimension Estimation

For accurate real-world measurements without any special equipment:

1. **Photo 1** — Stand at the **length side**, photograph across the width
2. **Photo 2** — Stand at the **breadth side**, photograph along the length

```python
from infer import load_model, estimate_dimensions_dual
import cv2

model      = load_model("checkpoints/best.pt")
length_img  = cv2.imread("from_length_side.jpg")
breadth_img = cv2.imread("from_breadth_side.jpg")

dims = estimate_dimensions_dual(model, length_img, breadth_img)
# {
#   "length_m":  5.2,
#   "breadth_m": 2.6,
#   "area_m2":   13.52
# }
```

---

## 💰 Pricing Logic

| Type | Formula | Default |
|---|---|---|
| Hourly | Fixed | ₹50 |
| Daily | Fixed | ₹300 |
| Monthly | `area_m² × ₹10` | e.g. 15m² → ₹150 |

All prices are configurable — pass custom values to `calculate_pricing()`.

---

## 🏋️ Training Details

| Setting | Value |
|---|---|
| Dataset | 5,000 synthetic images (OpenCV generated) |
| Train / Val split | 85% / 15% |
| Optimizer | AdamW (`lr=1e-3`, `weight_decay=1e-4`) |
| Scheduler | Cosine Annealing |
| Phase 1 (epochs 1–10) | Backbone frozen, head only |
| Phase 2 (epochs 11–30) | Last 5 backbone layers unfrozen |
| Augmentations | Color jitter, shadows, noise, blur, brightness |
| Best val corner error | < 10px on 224×224 images |

### Training Curves

Training takes ~10 minutes on a free Colab T4 GPU.

---

## 📦 Requirements

```
torch>=2.0.0
torchvision>=0.15.0
opencv-python-headless>=4.7.0
numpy>=1.24.0
tqdm>=4.65.0
matplotlib>=3.7.0
Pillow>=9.5.0
huggingface_hub>=0.16.0
```

Install all:
```bash
pip install -r requirements.txt
```

---

## 🔮 Roadmap

- [ ] Fine-tune on real parking space photos
- [ ] Add support for non-rectangular spaces (polygons)
- [ ] ONNX export for browser-based inference
- [ ] Integrate with Parkospace owner dashboard
- [ ] Occupied / available space detection
- [ ] Multi-space detection in a single image

---

## 🤝 Contributing

Contributions are welcome! If you have real parking space photos you'd like to contribute to improve the model, please open an issue or pull request.

1. Fork the repository
2. Create your feature branch: `git checkout -b feature/my-feature`
3. Commit your changes: `git commit -m 'Add my feature'`
4. Push to the branch: `git push origin feature/my-feature`
5. Open a Pull Request

---

## 📄 License

```
MIT License

Copyright (c) 2025 Parkospace

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
```

---

## 👤 Author

**UmeshAdabala** — [Parkospace](https://github.com/ParkoSpace/in)

> Built with ❤️ for making parking smarter in India 🇮🇳