This repo provides the implementation of the following ICCV 2021 paper: "Multimodal Knowledge Expansion" (MKE). Paper, Project Page.
We generate sythetic TwoMoon data. The data located at the upper moon and lower moon have true labels 0 and 1, and are colored by red and blue, respectively. Two modalities (α and β) are interpreted as observing from the X-axis and Y-axis.
Code: see TwoMoon/mke.ipynb
Image + Audio emotion recognition
Dataset: The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS). [link]