One-Shot Object Localization Using Learnt Visual Cues via Siamese Networks

TL;DR

This paper presents a neural network approach using Siamese networks to enable robots to recognize and localize novel objects in unstructured environments based on visual cues, demonstrated through simulation and datasets.

Contribution

Introduces an end-to-end Siamese network model for one-shot object localization using visual cues, adaptable to new, unseen objects in robotic applications.

Findings

Successful localization of novel objects in simulation

Effective use of visual cues like laser pointers for object identification

Good performance on Omniglot and toy datasets

Abstract

A robot that can operate in novel and unstructured environments must be capable of recognizing new, previously unseen, objects. In this work, a visual cue is used to specify a novel object of interest which must be localized in new environments. An end-to-end neural network equipped with a Siamese network is used to learn the cue, infer the object of interest, and then to localize it in new environments. We show that a simulated robot can pick-and-place novel objects pointed to by a laser pointer. We also evaluate the performance of the proposed approach on a dataset derived from the Omniglot handwritten character dataset and on a small dataset of toys.