Learning Condition-Invariant Scene Representations for Place Recognition across the Seasons Using Auto-Encoder and ICA

Figure 2

First column: two images from the same locations in two different seasons (Nordland dataset). Second column: images are projected on the first 100 principal components (PCs). Third column: images are projected on the second 100 main components (PCs); dependent features of the scenes are associated with the first few principal components.