: Adversarial machine learning, specifically targeting semantic segmentation networks (e.g., PSPNet, ICNet).

You can often find these supplementary videos on platforms like arXiv (under the "Ancillary files" section) or the researchers' project GitHub repositories.

: Files like "apns-218.mp4" typically show a side-by-side comparison of: The original input video. The adversarial patch being applied to the scene.

The resulting produced by the neural network.

: The authors demonstrate that a small patch placed in a scene can cause a segmentation model to fail globally or ignore critical objects (like pedestrians or traffic signs).