caffe随笔
(1)Blob数据类型:
存储data时: num*channel*height*width
存储conv layer时: H(output)*K(input)*height*width
(2)cpu_data与cpu_mutable_data的区别
cpu_data:只读取数据时用这个
cpu_mutable_data:有数据修改时用这个
(3)常用到的各种loss的类型:
分类:SOFTMAX_LOSS,HINGE_LOSS
线性回归:EUCLIDEAN_LOSS
Attributes /Multiclassification:SIGMOID_CROSS_ENTROPY_LOSS
SoftmaxLoss Layer
Used forpredicting a single class of K mutaully exclusice classes
Sigmoid cross-entropy loss
Binary logistic regression :used forpredicting K independent probability values in [0,1]
Euclidean Loss
A loss for regressing to real-valued labels[-inf,inf]
同时也可以采用多中loss,将不同的losslayer并联