一、deploy.prototxt文件
跟train.prototxt有一些差别,data层和softmax层做了修改
1、data层:删掉原来的数据层,换成一下代码
name: "CaffeNet"
layer {name: "data"
type: "Input"
top: "data"
input_param { shape: { dim: 1 dim: 3 dim: 63 dim: 63 } }
}
2、softmax层:删掉accuracy和softmaxwithloss层,换成一下代码
layer {
name: "prob"type: "Softmax"
bottom: "fc8"
top: "prob"
}
二、均值文件mean.binaryproto转成mean.npy文件
import numpy as np
import sys,caffe
root='/home/ths/6_operation_32_data/'
mean_proto_path=root+'6_operation_32_nearest2_test_lmdbmean.binaryproto'
mean_npy_path=root+'6_operation_32_nearest2_test_lmdbmean.npy'
blob=caffe.proto.caffe_pb2.BlobProto()
data=open(mean_proto_path,'rb').read()
blob.ParseFromString(data)
array=np.array(caffe.io.blobproto_to_array(blob))
mean_npy=array[0]
np.save(mean_npy_path,mean_npy)
三、synset_words.txt
这个文件是标签和对应的索引值,要与训练过程时的标签和索引保持一致
0 ORIGINAL