Fin-tuning....
●Rename the layers you need tochange.
●
Call CopyTrainedLayerFrom(pretrained_file)
●
Train as normal
layer{
layer {
name: “fc8”
name: “fc8-t”
type: “innerproduct”
type: “innerproduct”
num_output: 1000
num_output: 397
… …}
}
扫一扫
3038
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
