zongyuange / mixdcnn Goto Github PK

hello, i am reading your paper . now i know about process of the test with regard to the mixDCNN, but i can not find the train process in the paper. or maybe i was too careless , so can u give me some advice.

1. the code does not compile with cudnn v4
2. after set cudnn 0, got following error :
   examples/cifar10/convert_cifar_data.cpp:18:29: fatal error: caffe/util/db.hpp: No such file or directory
   #include "caffe/util/db.hpp"

Any idea how to fix this ?

For the "GoogleNet_birdsnap_6.prototxt" you used some kind of multi-label since you duplicate layers for each - prob-1, prob-2, sum-weighted-prob-1, sum-weighted-prob-2 and so on.

your "generate_caffe_mixture_dcnn_prototxt.py" can not generate this kind of network.
Could you please describe why is birdsnap has this type of network and how did you create it?