Industrial Smoke Emission Images with Pixel Masks

IJmond Industrial Smoke Segmentation Dataset contains 900 raw images and 2074 cropped images for pixel-level classification. The collection includes 1209 annotated smoke polygons and 1109 cropped images with corresponding pixel-level masks. It was created by Y.C. Hsu and is hosted on figshare.

Use Cases

• Train a semantic segmentation model on pixel-level masks to detect industrial smoke emissions in raw images.
• Use the 1209 annotated smoke polygons for polygon-based instance segmentation of smoke plumes.
• Fine-tune a computer vision model on 1109 cropped images with smoke masks for environmental monitoring tasks.
• Analyze the 7 raw images without smoke as negative samples to improve model robustness.

Strengths

• Includes 900 raw images and 2074 cropped images, providing a medium-scale dataset for segmentation tasks.
• Offers 1209 manually annotated smoke polygons and 1109 pixel-level masks for precise model training.
• Contains 7 raw images without smoke, which can serve as negative examples for model evaluation.

Limitations

• The dataset is relatively small for deep learning, with only 900 raw images and 2074 cropped images.
• Geographic scope is limited to the IJmond area, which may not generalize to other industrial regions.
• No temporal metadata is provided, limiting time-series or trend analysis applications.

Provenance

Source

Y.C. Hsu via figshare

Collection Method

Images and annotations collected for industrial smoke segmentation research.

Freshness

Last updated on March 24, 2026.

Geography

IJmond region (implied by dataset name).