Cigarette Butt Dataset & Trained Weights


This dataset consists of a set of 2200 synthetically composed images of cigarettes on the ground. It is designed for training CNNs (convolutional neural networks).

  • Annotations: Segmented, object-detection COCO format with custom categories. For more info, check out
  • Composition: Images were composed automatically with custom code utilizing the Python Imaging Library to apply random scale, rotation, brightness, etc to the foreground cutouts
  • Location: Photos of the ground and cigarette butts were taken in Austin, Texas
  • Camera: iPhone 8, original pixel resolution 3024 x 4032

Trained Weights

The trained weights were created with this source code. (You can download the weights separately from the dataset)

More Info

Immersive Limit Tutorial: Using Mask R-CNN with a Custom COCO-like Dataset

License Agreement & Download

You must read and accept the terms of the Non-Commercial, Educational License Agreement to download and use this content.

In essence it says:

  • You may only use the content for non-commercial research and educational purposes.
  • If you share the content, you must attribute Adam Kelly and include a link to (or a specific sub-page)
  • There is no warranty or liability
License Agreement *