Training schedule

[leochli]

Hi @camel007 ,

ShuffleNet is a great idea and the model size is apparently reduced.

Have you ever tried to train it on imagenet?

I only got 64% as top1 acc (1x, group = 3). My training schedule is 100 epochs, with batchsize 256, lr rate starts from 0.1 and decay by 0.1 every 30 epochs.

I'd be appreciated if there's any tricks in training to reach the paper acc.

thx!

[LiuRJun]

hi，I tried to train the ShuffleNet on my own database, but it always got a wrong and I do not know how to deal with it, can you share you train.prototxt, please?

[LiuRJun]

[libprotobuf ERROR google/protobuf/text_format.cc:245] Error parsing text-format caffe.NetParameter: 62:21: Message type "caffe.LayerParameter" has no field named "scale_param".

That the error display, do you know the reason ? If somewhere need to change ?

[miaomi1994]

@LiuRJun did you solve this problem?