Lagrangian Motion Magnification with Landmark-Prior and Sparse PCA for Facial Microexpressions and Micromovements.

Video motion magnification methods are motion visualization techniques that aim to magnify subtle and imper-ceptibly small motions in videos. They fall into two main groups where Eulerian methods work on the pixel grid with implicit motion information and Lagrangian methods use explicitly estimated motion and modify point trajectories. The motion in high framerate videos of faces can contain a wide variety of information that ranges from microexpressions over pulse or respiratory rate to cues on speech and affective state.

Detecting failure modes in image reconstructions with interval neural network uncertainty.

Purpose: The quantitative detection of failure modes is important for making deep neural networks reliable and usable at scale. We consider three examples for common failure modes in image reconstruction and demonstrate the potential of uncertainty quantification as a fine-grained alarm system.

Methods: We propose a deterministic, modular and lightweight approach called Interval Neural Network (INN) that produces fast and easy to interpret uncertainty scores for deep neural networks.