search

willard-yuan / cnn-for-image-retrieval

:sunrise:The code of post "Image retrieval using MatconvNet and pre-trained imageNet"
633 stars 227 forks source link

retrieval_virsulazation.m建议计算score用矩阵相乘,图像几十万张的时候会快很多 #6

Closed piaobuliao closed 7 years ago

piaobuliao commented 7 years ago

tools/retrieval_virsulazation.m

% %for loop = 1:n % % VecTemp = featNorm(loop, :); % % score(loop) = QueryVec*VecTemp'; % %end

score = (QueryVec*featNorm')';

用了http://www.vlfeat.org/matconvnet/pretrained/下的vgg-m-1024模型提取特征,CPU差不多一秒钟5张,GPU(GTX970)快了不少一秒钟50张左右

willard-yuan commented 7 years ago

@piaobuliao 用矩阵相乘代替循环确实会加速计算，已经在相应的地方修改，非常谢谢指出。