LiJianfei06 / MnasNet-caffe

A caffe implementation of Mnasnet: MnasNet: Platform-Aware Neural Architecture Search for Mobile.
52 stars 27 forks source link

精度问题 #5

Open Lucas-work opened 5 years ago

Lucas-work commented 5 years ago

你好,我用自己的猫狗数据来做训练,发现精度一直上不去。 而且训练速度很慢。

我没有自己写depthwise conv, 只用了caffe提供的group 操作。

另外我的batch size 设置到了128

请问这个问题应该则怎么解决