Learned Parametric Emotion Editing: Real-Time Affective Filtering for On-Device Social Media Video
A model that predicts this transformation in a single forward pass is learned: a MobileNetV4 backbone with FiLM-based emotion conditioning outputs parameters for differentiable global transformations, which reduced viewer-reported arousal relative to unedited images, comparably to the grayscale well-being filter.
Musa Rochi, Marcel Schubert, Christoph Gebhardt
· 0 citations