https://github.com/BVLC/caffe/blob/dev/src/caffe/proto/caffe.proto
这个是要参考的,里面有各种参数的说明。
layer {
name: "car_detector"
type: "Data"
top: "data"
top: "label"
include {
phase: TRAIN
}
transform_param {
mean_file: "../train_set/car_detector_mean.binaryproto"
scale: 0.00390625
}
data_param {
source: "../train_set"
batch_size: 64
backend: LMDB
mirror: true
}
}
这个里面,mean_file 和 scale都要放在transform_param中,虽然caffe.proto中也有规定data_param中也可以,但是实测不可以,也许是transform_param的存在导致失败。
以防万一,以后都统一放在tran。。。中吧。
src/caffe/proto/caffe.proto
You can also write your own solver - check out caffe/src/caffe/solver.cpp for details.
To convert the data, caffe/tools/extra/ may contain multiple scripts that are useful.