caffe Python API 之 數據輸入層（Data,ImageData,HDF5Data）

import sys
sys.path.append('/projects/caffe-ssd/python')
import caffe
4 net = caffe.NetSpec()

一、ImageData Layer

net.data ,net.label = caffe.layers.ImageData(
    name="InputData"
    source="train.txt",
    batch_size=32,
    new_width=48,
    new_height=48,
    ntop=2,
    is_color=True,
    shuffle=True,
    root_folder='/',
    transform_param=dict(crop_size=40,mirror=True))
print str(net.to_proto())

輸出：
layer {
  name: "InputData"
  type: "ImageData"
  top: "data"
  top: "label"
  transform_param {
    mirror: true
    crop_size: 40
  }
  image_data_param {
    source: "train.txt"
    batch_size: 32
    shuffle: true
    new_height: 48
    new_width: 48
    is_color: true
    root_folder: "/"
  }
}

 

二、Data Layer (lmdb/leveldb)

net.data, net.label = caffe.layers.Data(
    name="InputData",
    source="train_lmdb",
    backend = caffe.params.Data.LMDB,
    batch_size=32,
    ntop=2,
    include=dict(phase=caffe.TRAIN)
    transform_param=dict(
        crop_size=227,
        mean_value=[104, 117, 123],
        mirror=True
    )
)

輸出：
layer {
  name: "InputData"
  type: "Data"
  top: "data"
  top: "label"
  transform_param {
    mirror: true
    crop_size: 227
    mean_value: 104
    mean_value: 117
    mean_value: 123
  }
  data_param {
    source: "train_lmdb"
    batch_size: 32
    backend: LMDB
  }
}

三、HDF5Data Layer

net.data, net.label = caffe.layers.HDF5Data(
            name="InputData",
            source='./training_data_paths.txt',
            batch_size=64,
            include=dict(phase=caffe.TRAIN),
            ntop=2
            )

輸出：
layer {
  name: "InputData"
  type: "HDF5Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  hdf5_data_param {
    source: "./training_data_paths.txt"
    batch_size: 64
  }
}

另有：
image = L.HDF5Data( 
    hdf5_data_param={ 'source': './training_data_paths.txt', 'batch_size': 64 },
     include={'phase': caffe.TRAIN } 
)