Canada Guesser
Welcome to Canada Guesser, a collection of resources for building and exploring Canadian city classification models using street-view imagery. This organization contains a dataset, models, and an interactive demo designed to work together, enabling end-to-end experimentation and deployment.
Dataset
- Purpose: Train and evaluate models that classify street-view images by Canadian city.
- Content: 135,000 training images, 15,000 test images, covering 15 major Canadian cities with 10,000 images each.
- Features:
image, label (0–14), city (city name).
- Known Limitations: Some cities (Saskatoon, Halifax) contain dashcam-style images including dashboards, which may affect model behavior.
- License: CC-BY-SA-4.0
Model
- Purpose: Classifies Canadian street-view images by city.
- Trained on: Canadian Street View Cities dataset.
- Architecture: CNN (ConvNeXt-tiny) and Transformer (SwinV2) variants available.
- Task: Image classification.
Model Performance
| Model |
Accuracy |
Macro Precision |
Macro Recall |
Macro F1-Score |
| ConvNeXt-tiny |
0.98980 |
0.98983 |
0.98980 |
0.98980 |
| Swin Transformer V2 |
0.99440 |
0.99439 |
0.99440 |
0.99439 |
Performance was evaluated on the test split of the dataset. Both models achieve high accuracy, with Swin Transformer V2 slightly outperforming ConvNeXt-tiny.
Space
- Interactive app: Upload a street-view image and see which city the model predicts.
- Demonstrates the full pipeline: dataset → model → live inference.
Citation
If you use this dataset or models, please cite:
- Stephen Rebel, Danial McIntyre, Sharav Bali. Canadian Street View Classifier. Hugging Face, 2025.