search

FengyangZhang / caffe_roomnet

Segnet-like CNN for room layout estimation
Other
27 stars 7 forks source link

Data Preprocessing #6

Open kwangHouzz opened 7 years ago

kwangHouzz commented 7 years ago

Hi Fengyang,

How do you do data preprocessing (like scaling, subtracting mean value) in your model? It seems that you do not specify any scaling factor or mean value in either roomnet_multistage.prototxt or roomnet_train.prototxt?

Thanks!

FengyangZhang commented 7 years ago

Those are in the roomnet_data_layer.cpp I think.

kwangHouzz commented 7 years ago

Yes I understand that data transformation is defined in roomnet_data_layer.cpp. But the problem is that you are not defining any value of scaling factor/ mean in the prototxt. Was is true that you were not doing any data preprocessing in your training/testing? Please correct me if I am wrong.

Thanks!

kwangHouzz commented 7 years ago

BTW, what accuracy does your type classifier obtain in your experiment? I tried to train your model, and the validation accuracy was not even close to that of the original paper (81.5%).