In this paper, we discuss the adaptation of our decentralized place recognition method described in  to fullimage descriptors. As we had shown, the key to making a scalable decentralized visual place recognition lies in exploting deterministic key assignment in a distributed key-value map. Through this, it is possible to reduce bandwidth by up to a factor of n, the robot count, by casting visual place recognition to a key-value lookup problem. In , we exploited this for the bagof-words method , . Our method of casting bag-of-words, however, results in a complex decentralized system, which has inherently worse recall than its centralized counterpart. In this paper, we instead start from the recent full-image description method NetVLAD . As we show, casting this to a key-value lookup problem can be achieved with k-means clustering, and results in a much simpler system than . The resulting system still has some flaws, albeit of a completely different nature: it suffers when the environment seen during deployment lies in a different distribution in feature space than the environment seen during training.
- Detailed record: https://infoscience.epfl.ch/record/232969?ln=en