search

raghakot / keras-resnet

Residual networks implementation using Keras-1.0 functional API
Other
1.39k stars 616 forks source link

add BatchNormalization layers after 1 * 1 convolution in the shortcut path #12

Closed scihacker closed 7 years ago

scihacker commented 8 years ago

In the official version keras.datasets.resnet50, the 1 * 1 convolution layer on the shortcut path is continued with a batchnormalization layer. I tried the network without this layer with CIFAR-10 dataset, and it seems the response of convolution path is overwhelmed by the shortcut path. In this way, training has to be very slow, then reach a low accuracy plateau.

raghakot commented 7 years ago

It's definitely possible. The intent of this module was to provide a quick way to build your own resnets with the resnet-50 as a sample. By no means is that a fixed architecture.