Road Segmentation Model
Overview
This model performs road segmentation from satellite imagery using a deep learning semantic segmentation architecture.
The model takes an RGB satellite image as input and predicts a binary road mask indicating road locations.
Model Details
Architecture
- Framework: PyTorch
- Deployment Format: TorchScript (.pt)
- Task: Semantic Segmentation
- Output: Binary road mask
Input
- RGB satellite image
- Expected shape: 3 × H × W
- Pixel values normalized to [0,1]
Output
- Single-channel road probability mask
- Higher values indicate higher confidence of road presence
Usage
import torch
model = torch.jit.load("road_model_Deployment.pt")
model.eval()
prediction = model(image_tensor)
Performance
| Metric | Score |
|---|---|
| IoU | Add Your Score |
| Dice Score | Add Your Score |
| Validation Loss | Add Your Score |
Limitations
- Performance may vary across geographic regions.
- Small roads may be difficult to detect.
- Results depend on image quality and resolution.
Intended Uses
Intended
- Road extraction
- Infrastructure mapping
- Transportation analysis
- Disaster response
Not Intended
- Safety-critical systems
- Autonomous vehicle control
- Real-time navigation
Files
- road_model_Deployment.pt
- inference.py
- requirements.txt
License
MIT License