Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,96 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: PM25Vision
|
| 3 |
+
tags:
|
| 4 |
+
- computer-vision
|
| 5 |
+
- pm2.5
|
| 6 |
+
- regression
|
| 7 |
+
- classification
|
| 8 |
+
- air-quality
|
| 9 |
+
- AQI
|
| 10 |
+
task_categories:
|
| 11 |
+
- image-classification
|
| 12 |
+
- other
|
| 13 |
+
license: cc-by-4.0
|
| 14 |
+
language:
|
| 15 |
+
- en
|
| 16 |
+
size_categories:
|
| 17 |
+
- 10K<n<100K
|
| 18 |
+
---
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
# PM25Vision
|
| 22 |
+
|
| 23 |
+
## Dataset Summary
|
| 24 |
+
PM25Vision (PM25V) is a large-scale dataset for estimating air quality (PM2.5) from street-level imagery. It pairs **Mapillary** photos with **World Air Quality Index (WAQI)** PM2.5 records, covering 2014–2025, 3,261 monitoring stations, and 11,114 cleaned and balanced images.
|
| 25 |
+
|
| 26 |
+
## Tasks
|
| 27 |
+
- **Regression**: Predict continuous PM2.5 values.
|
| 28 |
+
- **Classification**: Predict discrete AQI levels.
|
| 29 |
+
|
| 30 |
+
## Baseline Results
|
| 31 |
+
### Regression
|
| 32 |
+
| Model | R² | MAE | RMSE | Acc | F1 |
|
| 33 |
+
|-----------------|------|------|------|------|------|
|
| 34 |
+
| EfficientNet-B0 | 0.55 | 36.6 | 54.6 | 0.46 | 0.45 |
|
| 35 |
+
| ResNet50 | 0.50 | 38.6 | 57.5 | 0.44 | 0.35 |
|
| 36 |
+
| ViT-B/16 | 0.23 | 50.3 | 71.7 | 0.35 | 0.30 |
|
| 37 |
+
|
| 38 |
+
### Classification
|
| 39 |
+
| Model | Acc | F1 | Precision | Recall |
|
| 40 |
+
|-----------------|------|------|-----------|--------|
|
| 41 |
+
| ResNet50 | 0.44 | 0.38 | 0.48 | 0.37 |
|
| 42 |
+
| ViT-B/16 | 0.40 | 0.37 | 0.41 | 0.36 |
|
| 43 |
+
| EfficientNet-B0 | 0.40 | 0.34 | 0.42 | 0.33 |
|
| 44 |
+
|
| 45 |
+
## Dataset Structure
|
| 46 |
+
|
| 47 |
+
The dataset is organized into two main splits: **train** and **test**, each containing:
|
| 48 |
+
|
| 49 |
+
- **`images/`**: all image files used in the dataset.
|
| 50 |
+
- **`samples_by_bin/`**: a small set of 30 example images per AQI bin (for quick visual inspection).
|
| 51 |
+
- **`metadata.csv`**: a CSV file describing metadata (including pm2.5 labels) for each image.
|
| 52 |
+
|
| 53 |
+
### Metadata Fields
|
| 54 |
+
|
| 55 |
+
Each row in `metadata.csv` contains:
|
| 56 |
+
|
| 57 |
+
| Field | Type | Description |
|
| 58 |
+
|----------------|---------|--------------------------------------------------------------------------------------|
|
| 59 |
+
| **`image_id`** | int64 | Unique image identifier (from Mapillary). |
|
| 60 |
+
| `station_id` | int64 | WAQI monitoring station ID. |
|
| 61 |
+
| `captured_at` | object | Date when the image was captured (YYYY-MM-DD). |
|
| 62 |
+
| `camera_angle` | float64 | Camera orientation (if available). |
|
| 63 |
+
| `longitude` | float64 | Longitude of the station. |
|
| 64 |
+
| `latitude` | float64 | Latitude of the station. |
|
| 65 |
+
| `quality_score`| float64 | Image quality score from Mapillary (if available). |
|
| 66 |
+
| `downloaded_at`| object | Timestamp when the sample was downloaded. |
|
| 67 |
+
| **`pm25`** | float64 | Average PM2.5 value of the day that the image was captured(the label, in AQI value). |
|
| 68 |
+
| `filename` | object | Image filename, located in the `images/` directory. |
|
| 69 |
+
| `quality` | object | ResNet18 classified label for image quality (e.g., `good` or `bad`). |
|
| 70 |
+
| `pm25_bin` | object | Discrete AQI level label (e.g., `0–50`, `51–100`, etc.). |
|
| 71 |
+
|
| 72 |
+
### Splits
|
| 73 |
+
|
| 74 |
+
- **Train**: 80% of samples, balanced across AQI bins.
|
| 75 |
+
- **Test**: 20% of samples, balanced across AQI bins.
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
## Limitations
|
| 79 |
+
- WAQI temporal resolution is **daily**, may miss intra-day variation.
|
| 80 |
+
- Spatial accuracy limited to 5 km around stations.
|
| 81 |
+
- Rare extreme AQI classes remain underrepresented.
|
| 82 |
+
|
| 83 |
+
## Access
|
| 84 |
+
- Arxiv: ...
|
| 85 |
+
- Online demo: [pm25vision.com](http://www.pm25vision.com)
|
| 86 |
+
|
| 87 |
+
## Citation
|
| 88 |
+
```bibtex
|
| 89 |
+
@misc{pm25vision2025,
|
| 90 |
+
title = {PM25Vision: Street-level imagery with PM2.5 annotations},
|
| 91 |
+
author = {Han, Yang},
|
| 92 |
+
year = {2025},
|
| 93 |
+
publisher = {Hugging Face Datasets},
|
| 94 |
+
url = {https://huggingface.co/datasets/DeadCardassian/PM25Vision}
|
| 95 |
+
}
|
| 96 |
+
```
|