This is a list of awesome methods about data augmentation.
This library helps you with augmenting images for your machine learning projects. It converts a set of input images into a new, much larger set of slightly altered images. Many very popular projects have been integrated. New methods like augmix,cutmix，are being tracked. Whether you're a researcher or an engineer, just enjoy it!
Popular Projects
imgaug
- intro: 2019
- github star: 7.8k
- github: https://github.com/aleju/imgaug
Albumentations
Albumentations: fast and flexible image augmentations
- intro: ArXiv 2018
- github star: 4.1k
- arxiv: https://arxiv.org/abs/1809.06839v1
- github: https://github.com/albumentations-team/albumentations
Augmentor
Biomedical image augmentation using Augmentor
- intro: Bioinformatics
- github star: 3.7k
- arxiv: https://github.com/mdbloice/Augmentor
- github: https://github.com/mdbloice/Augmentor
- docs: https://augmentor.readthedocs.io/en/master/
Augmentor is a Python package designed to aid the augmentation and artificial generation of image data for machine learning tasks. It is primarily a data augmentation tool, but will also incorporate basic image pre-processing functionality.
Papers&Codes
mixup
Mixup: BEYOND EMPIRICAL RISK MINIMIZATION
- intro: ICLR2018
- arxiv: https://arxiv.org/abs/1710.09412
- github: https://github.com/facebookresearch/mixup-cifar10
Mixup is a generic and straightforward data augmentation principle. In essence, mixup trains a neural network on convex combinations of pairs of examples and their labels. By doing so, mixup regularizes the neural network to favor simple linear behavior in-between training examples.
Cutout
Improved Regularization of Convolutional Neural Networks with Cutout
- intro: arXiv 2017
- arxiv: https://arxiv.org/abs/1708.04552
- github: https://github.com/uoguelph-mlrg/Cutout
Cutmix
CutMix:Regularization Strategy to Train Strong Classifiers with Localizable Features
- intro: ICCV 2019 (oral talk)
- arxiv: https://arxiv.org/pdf/1905.04899.pdf
- github: https://github.com/clovaai/CutMix-PyTorch
Augmix
AUGMIX: A SIMPLE DATA PROCESSING METHOD TO IMPROVE ROBUSTNESS AND UNCERTAINTY
- intro: ICLR 2020
- arxiv: https://arxiv.org/pdf/1912.02781.pdf
- github: https://github.com/google-research/augmix
copy-paste
Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation
- intro: 2020
- provider: google
- arxiv: https://arxiv.org/pdf/2012.07177.pdf
- github: https://github.com/google-research/augmix
fast-autoaugment
Fast AutoAugment
- intro: NeurIPS 2019
- github star: 671
- arxiv: https://arxiv.org/abs/1905.00397
- github: https://github.com/kakaobrain/fast-autoaugment
AutoAugment
AutoAugment:Learning Augmentation Strategies from Data
- intro: CVPR 2019
- provider: google
- arxiv: https://arxiv.org/pdf/1805.09501v3.pdf
- github: https://github.com/DeepVoltaire/AutoAugment
RandAugment
RandAugment: Practical automated data augmentation with a reduced search space
- intro: ICLR 2020
- provider: google
- arxiv: https://arxiv.org/pdf/1909.13719.pdf
- github: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet
Random-Erasing
Random Erasing Data Augmentation
| black | white | random |
|---|---|---|
 |
 |
 |
 |
 |
 |
- intro: AAAI 2020
- arxiv: https://arxiv.org/pdf/1708.04896.pdf
- github: https://github.com/zhunzhong07/Random-Erasing
GridMask
GridMaskDataAugmentation
- intro: 2020.01
- arxiv: https://arxiv.org/abs/2001.04086
- github: https://github.com/akuxcw/GridMask
- 知乎参考: https://zhuanlan.zhihu.com/p/103992528
MMD
A Person Re-identification Data Augmentation Method with Adversarial Defense Effect
- intro: 2021.01
- arxiv: https://arxiv.org/abs/2001.04086
- github: https://github.com/finger-monkey/ReID_Adversarial_Defense
imagecorruptions
Benchmarking Robustness in Object Detection:Autonomous Driving when Winter is Coming
- intro: arXiv 2019
- arxiv: https://arxiv.org/abs/1807.01697
- github: https://github.com/CrazyVertigo/imagecorruptions
CycleGAN
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networkss
- intro: ICCV 2017
- arxiv: https://arxiv.org/pdf/1912.02781.pdf
- provider: UC Berkeley
- github: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix
- github: https://github.com/junyanz/CycleGAN
ALAE
Adversarial Latent Autoencoders
- intro: CVPR 2020
- arxiv: https://arxiv.org/pdf/2004.04467.pdf
- github: https://github.com/podgorskiy/ALAE
Small Object Augmentation
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networkss
- intro: 2017
- arxiv: https://arxiv.org/pdf/1902.07296.pdf
- github: https://github.com/gmayday1997/SmallObjectAugmentation
Segmentation & Matting
Real-Time High-Resolution Background Matting
- intro: 2020.12
- arxiv: https://arxiv.org/abs/2012.07810
- github: https://github.com/PeterL1n/BackgroundMattingV2
Image Composition:Deep Image Harmonization
Deep Image Harmonization via Domain Verification
- intro: CVPR 2020
- provider: SJTU
- arxiv: https://arxiv.org/abs/1911.13239
- github: https://github.com/bcmi/Image_Harmonization_Datasets
InstaBoost: Boosting Instance Segmentation Via Probability Map Guided Copy-Pasting
- intro: ICCV 2019
- provider: SJTU
- arxiv: https://arxiv.org/abs/1908.07801
- github: https://github.com/GothicAi/Instaboost
Hard data mining
Unsupervised Hard Example Mining from Videos for Improved Object Detection
- intro: ECCV 2018
- arxiv: http://vis-www.cs.umass.edu/unsupVideo/docs/unsup-video_eccv2018.pdf
- github: https://github.com/adiprasad/unsup-hard-negative-mining-mscoco
- project: http://vis-www.cs.umass.edu/unsupVideo/
- demo video: http://vis-www.cs.umass.edu/unsupVideo/docs/suppVideo.mp4
- 知乎参考: https://zhuanlan.zhihu.com/p/174057800
Annotation Tools
labelImg
- intro: 2017
- github star: 9.8k
- github: https://github.com/tzutalin/labelImg
LabelImg is a graphical image annotation tool and label object bounding boxes in images.
labelme
- intro: 2017
- github star: 4.2k
- github: https://github.com/wkentaro/labelme
Image Polygonal Annotation with Python (polygon, rectangle, circle, line, point and image-level flag annotation).
from
