The CMU CREATE Lab needs volunteers to help label smoke emissions from a collection of pollution monitoring videos! With sufficient labeled videos, we aim to train an Artificial Intelligence system to detect smoke emissions automatically. A brief introduction about this tool can be found at this video.

So far, [...] videos are fully labeled (confirmed by multiple users), and [...] videos are partially labeled. There are [...] videos in the dataset.

The system currently supports Android 7+, iOS 11+, and recent up-to-date desktop browsers. Before you start, note the following characteristics of smoke:

  • Smoke shows various colors, while steam is mostly whitish.
  • Smoke has unclear edges that fade away slower than steam.
  • Smoke has various opacities, while steam has extremely high opacity.

Please label high and low opacity smoke, even with unknown emission source. High-opacity smoke (the first and second videos below) blocks its background. For low-opacity smoke (the third and fourth videos below), the background can be visible.

Steam disappears faster and has sharp edges when compared to smoke. Steam also has extremely high opacity, which makes its background not visible. Here are examples that show mainly steam, which should not be selected.

Smoke and steam can appear at the same time. Here are examples that show both smoke and steam, which also needs to be selected.

Various bad weather conditions can affect the camera view. Here are examples that show bad weather, which should not be selected.

Please let us know which aspects are working well or could be improved via our 15-minute Survey Study. We appreciate your input!

Video autoplay is disabled. Please enable it.