Copy And Paste As Data Augmentation!

3 main points
✔️ Use of Copy-Paste algorithm for data augmentation in instance segmentation
✔️ Use of Copy-Paste algorithm in self-supervised (self-training) setting
✔️ New state of the art on COCO and LVIS datasets.

Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation
written by Golnaz Ghiasi, Yin Cui, Aravind Srinivas, Rui Qian, Tsung-Yi Lin, Ekin D. Cubuk, Quoc V. Le, Barret Zoph
(Submitted on 13 Dec 2020)
Comments: Accepted by arXiv.
Subjects: Computer Vision and Pattern Recognition (cs.CV)
  comm 

Introduction

Like most deep learning, training instance segmentation models is a highly data-intensive process. Moreover, it takes a considerable amount of time and manual work to prepare these datasets. Data augmentation can help reduce the time and resources required and increase data efficiency. This paper is based on one of such methods: Copy-Paste augmentation. Unlike most image augmentation techniques, this method is more object-aware, and therefore it is intuitive to assume that it would suit instance(object) segmentation well. In addition, this paper also studies the effects of Copy-Paste augmentation when used with self-training (i. e. in a semi-supervised learning approach). These techniques result in a significant improvement in the state of the art on the COCO and LVIS datasets.

Thapa Samrat

I am a second year international student from Nepal who is currently studying at the Department of Electronic and Information Engineering at Osaka University. I am interested in machine learning and deep learning. So I write articles about them in my spare time.