Self-Supervised Learning for Biologically-Inspired Place Representation Generalization across Appearance Changes
share
Visual place recognition is a key to unlocking spatial navigation for animals, humans and robots. From a biological perspective, research in neuroscience has shown that place recognition ability presupposes abstracting mental representations of places so that they can generalize to unseen conditions. Inspired by that, we investigate learning features that are insensitive to appearance modifications while sensitive to geometric transformations in a self-supervised manner. That dual-purpose training is made possible by combining the two self-supervision main paradigms, i.e. contrastive and predictive learning. Our results on standard benchmarks reveal that jointly learning such appearance-invariant and geometry-equivariant image descriptors leads to competitive visual place recognition results across adverse seasonal and illumination conditions, without requiring any human-annotated labels
READ FULL TEXT