[Paper] CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features (Image Classification)

CutMix: Patches are cut and pasted among training image

Outline

1. CutMix

2. Comparison with Cutout and mixup

Comparison among mixup, Cutout, and CutMix
Class activation mapping (CAM)

3. Experimental Results

3.1. Ablation Study

3.2. ImageNet

ImageNet classification results based on ResNet-50
ImageNet classification results based on ResNet-101 and ResNeXt-101

3.3. CIFAR

Error Rates on CIFAR-100
Impact of CutMix on lighter architectures on CIFAR-100.
Impact of CutMix on CIFAR-10.

PhD, Researcher. I share what I've learnt and done. :) My LinkedIn: https://www.linkedin.com/in/sh-tsang/, My Paper Reading List: https://bit.ly/33TDhxG