shicai / DenseNet-Caffe

DenseNet Caffe Models, converted from https://github.com/liuzhuang13/DenseNet
357 stars 273 forks source link

about some details #19

Closed dlyldxwl closed 7 years ago

dlyldxwl commented 7 years ago

Thanks for your job! I use your pretrain model and SSD algorithm to train a model. for conv layer, I add param { lr_mult: 1 decay_mult: 1} for bn layer , I add param { lr_mult: 0 decay_mult: 0} for 3 times to fix. I don't konw the scale layer need add lr_mult and decay_mult?scale 层需要添加学习率这些参数吗 And for conv layer, need I add weight_filler and bias_filler? 我既然有了预训练模型,那还需要对卷积层参数初始化吗? 谢谢!

shicai commented 7 years ago

1 需要,通常是2次{ lr_mult: 1 decay_mult: 0} 2 这个无所谓,无论初始化与否,都会被预训练模型的参数覆盖。

foralliance commented 6 years ago

@shicai 在微调时,   1. 如果某一层是预训练网络中的层,比如基于VGG16,conv5_3就是VGG16中的某一层,如果这一层出现了weight_filler和bias_filler这2个参数(初始化用的).那无论这2个参数是什么,都会被预训练模型的参数覆盖,   2. 如果某一层不是预训练网络中的层,那么weight_filler和bias_filler这2个参数就是用来初始化的.   可以这么理解吗??   麻烦了!!