CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features

ICCV 2019 Sangdoo YunDongyoon HanSeong Joon OhSanghyuk ChunJunsuk ChoeYoungjoon Yoo

Regional dropout strategies have been proposed to enhance the performance of convolutional neural network classifiers. They have proved to be effective for guiding the model to attend on less discriminative parts of objects (e.g. leg as opposed to head of a person), thereby letting the network generalize better and have better object localization capabilities... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification CIFAR-10 PyramidNet-200 + CutMix Percentage correct 97.12 # 23
Image Classification CIFAR-100 PyramidNet-200 + Shakedrop + Cutmix Percentage correct 86.19 # 10
Image Captioning COCO NIC (ResNet-50, CutMix) BLEU-1 64.2 # 1
BLEU-2 46.3 # 1
BLEU-3 33.6 # 1
BLEU-4 24.9 # 1
CIDEr 77.6 # 1
METEOR 23.1 # 1
ROUGE 49 # 1
Image Classification ImageNet ResNeXt-101 (CutMix) Top 1 Accuracy 81.53% # 43
Top 5 Accuracy 94.97% # 40
Image Classification ImageNet ResNet-50 (CutMix) Top 1 Accuracy 78.4% # 88
Top 5 Accuracy 94.10% # 62
Domain Generalization ImageNet-A CutMix (ResNet-50) Top-1 accuracy % 7.3 # 2
Object Detection PASCAL VOC 2007 SSD (CutMix) MAP 76.7% # 15
Object Detection PASCAL VOC 2007 SSD (ResNet-50, CutMix) MAP 77.6% # 13

Methods used in the Paper