Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation
Golnaz Ghiasi 1 Yin Cui 1 Aravind Srinivas 1,2 Rui Qian 1,3 Tsung-Yi Lin1 Ekin D. Cubuk1 Quoc V. Le1 Barret Zoph1 1Google Research Equal contribution. Correspondence to: done during an internship at Google Research. Brain Team 2 UC Berkeley 3 Cornell University
Abstract
Building instance segmentation models that are data-efficient and can handle rare object categories is an important challenge in computer vision. Leveraging data augmentations is a promising direction towards addressing this challenge. Here, we perform a systematic study of the Copy-Paste augmentation (\eg, [13, 12]) for instance segmentation where we randomly paste objects onto an image. Prior studies on Copy-Paste relied on modeling the surrounding visual context for pasting the objects. However, we find that the simple mechanism of pasting objects randomly is good enough and can provide solid gains on top of strong baselines. Furthermore, we show Copy-Paste is additive with semi-supervised methods that leverage extra data through pseudo labeling (\egself-training). On COCO instance segmentation, we achieve 49.1 mask AP and 57.3 box AP, an improvement of +0.6 mask AP and +1.5 box AP over the previous state-of-the-art. We further demonstrate that Copy-Paste can lead to significant improvements on the LVIS benchmark. Our baseline model outperforms the LVIS 2020 Challenge winning entry by +3.6 mask AP on rare categories. 111Code and checkpoints for our models are available at https:
中文速览
实例分割（instance segmentation）模型通常需要大量标注数据，而标注成本极高，如何用更少的数据训练出更好的模型是计算机视觉领域的核心挑战之一。研究者对一种叫做"复制粘贴"（Copy-Paste）的数据增强方法进行了系统研究，核心做法是：随机从一张图片中抠出目标物体，直接贴到另一张图片的随机位置，无需像以往方法那样费力建模周围环境的视觉上下文。实验表明，这种看似粗糙的随机粘贴策略配合大尺度抖动增强，在COCO基准上就能显著提升模型性能，数据利用效率提升约2倍，仅用10%训练数据时检测AP提升高达10个点；结合自训练（self-training）伪标签方法后，最终在COCO test-dev上达到49.1 mask AP和57.3 box AP的当时最优水平，在长尾数据集LVIS上稀有类别mask AP更超过2020年挑战赛冠军3.6个点。这项工作的意义在于：它用一个极其简单、零推理开销、即插即用的技巧，大幅刷新了实例分割的性能上限，为数据稀缺和类别不平衡问题提供了一条低成本的实用解法。
原文 arXiv:2012.07177;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2012.07177v2