我尝试使用leveldb通过caffe python接口制作train / val.prototxt:
layer {
name: "cifar"
type: "Data"
top: "data"
top: "label"
data_param {
source: "/home/youngwan/data/cifar10/cifar10-gcn-leveldb-splits/cifar10_full_train_leveldb_padded"
batch_size: 100
backend: LEVELDB
}
transform_param {
mean_file: "/home/youngwan/data/cifar10/cifar10-gcn-leveldb-splits/paddedmean.binaryproto"
mirror: 1
crop_size: 32
}
include: { phase: TRAIN }
}
但是在caffe python界面中,即使我试图在BLVC / caffe页面中找到示例和教程,我也找不到合适的数据层python包装器(例如,L.MemoryData
)。
您能注意到我可以使用哪个'L.xxx'
图层吗?
答案 0 :(得分:1)
使用caffe.NetSpec()
界面,您可以拥有所需的所有图层:
from caffe import layers as L, params as P
cifar = L.Data(data_param={'source': '/home/youngwan/data/cifar10/cifar10-gcn-leveldb-splits/cifar10_full_train_leveldb_padded',
'batch_size': 100,
'backend': P.Data.LEVELDB},
transform_param={'mean_file': '/home/youngwan/data/cifar10/cifar10-gcn-leveldb-splits/paddedmean.binaryproto',
'mirror': 1,
'crop_size': 32},
include={'phase':caffe.TRAIN})
基本上,L.<layer type>
定义了<layer type>
类型的图层。