Continual Learning of Object Instances

Kishan Parshotam, Mert Kilickaya

Student First Author: Yes

Previously Published: Continual Learning in Computer Vision (CLVISION CVPR Workshop 2020)

TL;DR: We propose continual instance learning by incrementally distinguishing instances of the same object category with metric learning.

Keywords: Continual Learning, Object Instances, Metric Learning, Re-Identification, Data Privacy

Abstract: We propose continual instance learning - a method that applies the concept of continual learning to the task of distinguishing instances of the same object category. We specifically focus on the car object, and incrementally learn to distinguish car instances from each other with metric learning. We begin our paper by evaluating current techniques. Establishing that catastrophic forgetting is evident in existing methods, we then propose two remedies. Firstly, we regularise metric learning via Normalised Cross-Entropy. Secondly, we augment existing models with synthetic data transfer. Our extensive experiments on three large-scale datasets, using two different architectures for five different continual learning methods, reveal that Normalised cross-entropy and synthetic transfer leads to less forgetting in existing techniques.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2004.10862/code)