1、Input layer

Input layer用在deploy文件测试模型效果，需要代码中手动指定网络输入数据，唯一的参数BlobShape设定输入数据的维度

caffe.proto中定义如下：

message InputParameter {
  // This layer produces N >= 1 top blob(s) to be assigned manually.
  // Define N shapes to set a shape for each top.
  // Define 1 shape to set the same shape for every top.
  // Define no shape to defer to reshaping manually.
  repeated BlobShape shape = 1;
}

prototxt文件定义如下：

layer {
    name:"data"    #设定当前layer名称为data
    type:"Input"   #设置当前layer类型为Input
    top:"data"     #设置当前layer输出blob名称为data
    #定义输入数据维度为 batchsize =1 channel=1 height=42 dim=42
    input_param {shape: {dim: 1 dim: 1 dim: 42 dim:42}}
}

python读取图片输入NET代码如下：

#创建神经网络
net = caffe.Net(net_file, weight_file, caffe.TEST)

#指定caffe数据转换器data的shape等于神经网络data层shape
transformer = caffe.io.Transformer({'data':net.blobs['data'].data.shape})

#指定交换维度，caffe读取的图片是H*W*C,交换后变成C*H*W
transformer.set_transpose('data', (2,0,1))    

#指定减去均值,np.load(mean_file)返回数据格式channel0数据：channel1数据：channel2数据
#mean(1).mean(1)返回每个通道的均值
transformer.set_mean('data', np.load(mean_file).mean(1).mean(1))

#caffe.io.load_image()读进来是RGB格式和（0，1）(float）
#cv2.imread()读进来直接是BGR格式和（0，255）
#这里是将RGB变换到BGR
transformer.set_chanenl_swap('data', (2,1,0))
#这里是将图像数据变成(0,255),注释掉该行代码表示使用(0,1)的数据
transformer.set_raw_scale('data',255)

#读取图片
im=caffe.io.load_image(image_path, 0) 
#转换器转换数据，并赋值给net的data层
net.blobs['data'].data[...] = transformer.preprocess('data', im)
out = net.forward()

2、Data Layer层

指定Net中输入数据为LEVELDB或者LMDB

message DataParameter {
  enum DB {
    LEVELDB = 0;
    LMDB = 1;
  }
  //------------这些都是data层本身的参数设定data_param
  // Specify the data source.
  //指定数据文件路径
  optional string source = 1;    
  // Specify the batch size.
  //指定batch size 大小
  optional uint32 batch_size = 4;
  // The rand_skip variable is for the data layer to skip a few data points
  // to avoid all asynchronous sgd clients to start at the same point. The skip
  // point would be set as rand_skip * rand(0,1). Note that rand_skip should not
  // be larger than the number of keys in the database.
  // DEPRECATED. Each solver accesses a different subset of the database.
  optional uint32 rand_skip = 7 [default = 0];
  //指定数据类型，默认是LEVELDB
  optional DB backend = 8 [default = LEVELDB];
 

  //------之后是数据增强的参数设置，transform_param
  // DEPRECATED. See TransformationParameter. For data pre-processing, we can do
  // simple scaling and subtracting the data mean, if provided. Note that the
  // mean subtraction is always carried out before scaling.
  optional float scale = 2 [default = 1]; 
  //设置均值文件路径
  optional string mean_file = 3;
  
  // DEPRECATED. See TransformationParameter. Specify if we would like to randomly
  // crop an image.
  //指定裁剪后图片的大小
  optional uint32 crop_size = 5 [default = 0];
  // DEPRECATED. See TransformationParameter. Specify if we want to randomly mirror
  // data.
  //指定是否通过镜像座扩展
  optional bool mirror = 6 [default = false];
  // Force the encoded image to have 3 color channels
  optional bool force_encoded_color = 9 [default = false];
  // Prefetch queue (Increase if data feeding bandwidth varies, within the
  // limit of device memory for GPU training)
  optional uint32 prefetch = 10 [default = 4];
}

prototxt文件定义如下：

layer {
    name:"data"
    type:"Data"
    top:"data"
    top:"label"
    include {
        phase: TRAIN
    }
    transform_param {
        mirror: true
        crop_size: 42
        mean_file: "/home/zxy/caffe/face_expression_recognition/lmdb/meantrain.binaryproto"
    }
    data_param {
        source: "/home/zxy/caffe/face_expression_recognition/lmdb/fer2013_train_lmdb"
        batch_size: 128
        backend: LMDB
    }
}

Caffe Layer 系列(一)：Input层、Data层

1、Input layer

2、Data Layer层

猜你喜欢