caffenet-benchmark/batchnorm.md at master · ducha-aiki/caffenet-benchmark

This is quick evaluation of BatchNorm layer (BVLC/caffe#3229) performance on ImageNet-2012. Other on-going evaluations: activations [architectures] (https://github.com/ducha-aiki/caffenet-benchmark/blob/master/Architectures.md) [augmentation] (https://github.com/ducha-aiki/caffenet-benchmark/blob/master/Augmentation.md) The architecture is similar to CaffeNet, but has differences: Images are resiz

[stealthinu]

こっちの実験だとReLU-BNのほうが、論文に書かれてたBN-ReLUのパターンよりも良いと書かれている。あとBNと弱いDropout(0.2)併用の結果がよくなることも示されてる。

[yu4u]

batch normalizationとactivationの順番とかdropoutとの組み合わせとかの評価