当前位置: 代码迷 >> 综合 >> 【数据增强】CutMix: Regularization Strategy to Train Strong Classi?ers with Localizable Features
  详细解决方案

【数据增强】CutMix: Regularization Strategy to Train Strong Classi?ers with Localizable Features

热度:50   发布时间:2024-01-28 23:25:03.0

x是图片,y是图片对应的label,通过合成两张训练图片a和b,生成新的训练样本。

M是二值mask,大小与原图片一致。lam来源于Beta(alp, alp)分布。Alp设置为1,lam取自均匀分布(0,1)。设置bbox B,它的坐标是:

裁剪图片a中的区域B,用图片b中的区域B填充。Mask M的长宽比例和原图片的一致。Bbox坐标通过下式得到:

具体算法如下:

.

结果:

On ImageNet classi?cation, applying CutMix to ResNet-50 and ResNet101 brings +2.28% and +1.70% top-1 accuracy improvements. On CIFAR classi?cation, CutMix signi?cantly improves the performance of baseline by +1.98% leads to the state-of-the-art top-1 error 14.47%

  相关解决方案