Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation

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 ([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 (e.g. self-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.

PDF Abstract CVPR 2021 PDF CVPR 2021 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Object Detection COCO minival Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) box AP 57.0 # 41
Instance Segmentation COCO minival Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) mask AP 48.9 # 28
Object Detection COCO minival Cascade Eff-B7 NAS-FPN (1280) box AP 54.5 # 53
Instance Segmentation COCO minival Cascade Eff-B7 NAS-FPN (1280) mask AP 46.8 # 40
Instance Segmentation COCO test-dev Cascade Eff-B7 NAS-FPN (1280) mask AP 46.9 # 32
Object Detection COCO test-dev Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) box mAP 57.3 # 34
Object Detection COCO test-dev Cascade Eff-B7 NAS-FPN (1280) box mAP 54.8 # 48
Instance Segmentation COCO test-dev Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) mask AP 49.1 # 22
Object Detection LVIS v1.0 val Eff-B7 NAS-FPN (1280, Copy-Paste pre-training)) box AP 41.6 # 9
Instance Segmentation LVIS v1.0 val Eff-B7 NAS-FPN (1280, Copy-Paste pre-training)) mask AP 38.1 # 9
Object Detection PASCAL VOC 2007 Cascade Eff-B7 NAS-FPN (Copy Paste pre-training, single-scale) MAP 89.3% # 1
Semantic Segmentation PASCAL VOC 2012 val Eff-B7 NAS-FPN (Copy-Paste pre-training, single-scale)) mIoU 86.6% # 3

Methods