Transferable Feature Representation for Visible-to-Infrared Cross-Dataset Human Action Recognition

Figure 9

Examples of the three first dimensions of the original, aligned, and generalized data space. The left panel shows data distributions of two domains per class while the right panel shows the data distributions of all classes per domain.

(a) Original data

(b) Aligned data by KEMA

(c) Generalized data by aligned-to-generalized encoders