随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling

S3Pool: Pooling with Stochastic Spatial Sampling
CVPR2017
https://github.com/Shuangfei/s3pool

本文将常规池化看作两个步骤:
1)以步长为1在特征图上滑动池化窗口,尺寸大小基本保持不变, leaves the spatial resolution intact
2)以一种 uniform 和 deterministic 的方式进行降采样
我们认为这种 uniform 和 deterministic 的降采样方式 对于学习来说不是最优的( which aims for generalization to unseen examples ),对此我们提出一种随机采样方式,这种随机采样 像一个 strong regularizer,可以被看作 implicit data augmentation by introducing distortions in the feature maps

Stochastic pooling 可以看作在一个池化窗口内 对特征图数值进行归一化, 按照特征图归一化后的 概率值大小随机采样选择,即元素值大的被选中的概率也大

随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第1张图片
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第2张图片
g being a hyperparameter named grid size

通过改变 grid size g 来控制 distortion/stochasticity
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第3张图片

测试用的网络结构
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第4张图片

NIN on CIFAR-10 and CIFAR-100
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第5张图片

ResNet on CIFAR-10 and CIFAR-100
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第6张图片

Effect of grid size
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第7张图片
通过增加grid size 尺寸,训练误差变大,对应更多的随机性。测试误差先降低( stronger regularization),后来升高(当训练误差太高时)。

This observation suggests a trade-off between the optimization feasibility and the generalization ability, which can be adjusted in different applications by setting the grid sizes of each S3Pool layer

Results on STL-10
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第8张图片

训练样本数量和测试误差的关系
随机采样池化--S3Pool: Pooling with Stochastic Spatial Sampling_第9张图片

你可能感兴趣的:(CVPR2017,CNN网络结构和模型)