Hi,

I am using caffe to extract CNN features from pretrained model by imagenet. Everything goes smooth. But I find a problem with C++ code that, if I use C++ code to do the feature extraction as instructed in http://caffe.berkeleyvision.org/gathered/examples/feature_extraction.html, the features are a lot different than features extracted by python and matlab wrappers( these two wrappers give same results!) and the C++ extracted features have a lot lower accuracy on my test samples...

Do you have any comments or suggestions? many thanks!