szagoruyko / cvpr15deepcompare

Code and models for "Learning to Compare Image Patches via Convolutional Neural Networks"
http://imagine.enpc.fr/~zagoruys/deepcompare.html
Other
468 stars 169 forks source link

how to use the caffe model? #8

Open tingtinglu opened 8 years ago

tingtinglu commented 8 years ago

I am new to deep learning, and i only know how to use caffe a little. If i want to use the caffe model to test the pre-trained model, how should i preprocess the input images? or just give two images to the models? Thanks so much !

szagoruyko commented 8 years ago

you need to scale images to [0,1] range and subtract mean as in the example here https://github.com/szagoruyko/cvpr15deepcompare/blob/master/torch/extract_cpu.lua#L17

jeejeelee commented 7 years ago

Hi, Is your issue solved? using the caffe model of this repository, I can't get the correct output value([-1,1]) ,such as 10.111,-9.433

szagoruyko commented 7 years ago

@OpenPandaLee the outputs are not normalized to [-1,1] range by the network, the values you see are expected.

jeejeelee commented 7 years ago

@szagoruyko thanks for your help, so how do i evaluate similarity of images use the output?thank you very much