Air Quality (PM2.5) Prediction from Images
Upload a photo to predict the AQI category (model resizes to 224×224 internally).
Allowed types: JPG / PNG / WEBP • Max size: 10 MB
  • Large images will be compressed in your browser before upload.
  • We won’t upload files over 10 MB. Please choose a smaller file.
Preview will appear here
Prediction result will appear here.

AQI Reference Levels ↗

AQI Levels Reference

About this Project

This demo is powered by the PM25Vision Dataset, which combines Mapillary street-level imagery with air quality measurements from the World Air Quality Index (WAQI).

A deep learning model (EfficientNet-B0) was trained on this dataset to classify PM2.5 pollution levels. For more details, please visit our Kaggle dataset or Hugging Face dataset. See also our arXiv paper for technical details.

Note: When image quality is good, the model predicts the correct AQI class about 45% of the time. This demo is for research purposes only; no guarantees are provided and we are not responsible for any losses or misuse.