ENCP / CNNdroid

Open Source Library for GPU-Accelerated Execution of Trained Deep Convolutional Neural Networks on Android
MIT License
539 stars 181 forks source link

Confuse about some parameters #12

Closed hirorogithub closed 7 years ago

hirorogithub commented 7 years ago

When i writing net structure file about caffe-to-Cnndroid, i find some parameters that never appear in caffes net structure files but in CNNdroids,like normalization region,group.And i also confuse that i can`t find layer type like dropout layers. Could you tell what these parameters means?I'm afraid it was a bother for you to do this. I would be very greatful if you can give me a hand.

latifisalar commented 7 years ago

Hi, Group is used in convolution layer and in the AlexNet because of the multi-GPU tuning method they had proposed, you can just use a value of 1 for the other networks. Normalization region is also only used in LRN layer which is present in the AlexNet but not in the other CNNs. Dropout layer is used in the training phase, but CNNdroid is only meant to be used to run CNNs in the test phase, that's why it's not available in the library.

hirorogithub commented 7 years ago

Thanks for your help!