505-试验记录

说明：

• 记录试验过程以及数据
• 每次更新内容在最上，旧的内容在下，按照时间顺序排列

试验内容：

• 用迁移学习的方法，利用前期获得的视网膜血管分割 'caffemodel' ，在Caffe框架上进行finetune，重新得到 地面裂缝 识别的caffemodel
• 样例图片

  • 视网膜血管图片：

    
    
    

    视网膜血管.jpg

  • 地面裂缝图片：

    
    
    

    地面裂缝.jpg

2017-10-31

实验数据：

总览：

• retina_20171017_21_train_iter_250000.caffemodel (视网膜血管caffemodel)
• solver.prototxt
• train_val.prototxt
• train_val_deploy.prototxt

细览：

• solver.prototxt

net: "C:/Users/M&L/Documents/Python Scripts/TF/data/train_val.prototxt"
test_iter: 100
test_interval: 500
base_lr: 0.001
lr_policy: "step"
gamma: 0.1
stepsize: 40000
display: 20
max_iter: 10000
momentum: 0.9
weight_decay: 0.0005
snapshot: 30000
snapshot_prefix: "C:/Users/M&L/Documents/Python Scripts/TF/data/transfer_20171027_train"
solver_mode: GPU

• train_val.prototxt

name: "RetinaNet"
layer {
   type: "Data"
   top: "X" # same name as given in create_dataset!
   top: "y"
   include { 
    phase:TRAIN 
  }
  data_param {
    source: "C:/Users/M&L/Documents/Python Scripts/TF/data/train_lmdb"
    batch_size: 128
    backend: LMDB
  }
 }
layer {
   type: "Data"
   top: "X" 
   top: "y"
   include {
    phase: TEST
  }
  data_param {
    source: "C:/Users/M&L/Documents/Python Scripts/TF/data/test_lmdb"
    batch_size: 128
    backend: LMDB
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "X"
  top: "conv1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "conv1"
  top: "conv2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "conv2"
  top: "conv3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "conv4"
  type: "Convolution"
  bottom: "conv3"
  top: "conv4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4"
  type: "ReLU"
  bottom: "conv4"
  top: "conv4"
}
layer {
  name: "conv4_1"
  type: "Convolution"
  bottom: "conv4"
  top: "conv4_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_1"
  type: "ReLU"
  bottom: "conv4_1"
  top: "conv4_1"
}
layer {
  name: "conv4_2"
  type: "Convolution"
  bottom: "conv4_1"
  top: "conv4_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_2"
  type: "ReLU"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "fc5-my"
  type: "InnerProduct"
  bottom: "conv4_2"
  top: "fc5-my"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5"
  type: "ReLU"
  bottom: "fc5-my"
  top: "fc5-my"
}
layer {
  name: "drop5"
  type: "Dropout"
  bottom: "fc5-my"
  top: "fc5-my"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc6"
  type: "InnerProduct"
  bottom: "fc5-my"
  top: "fc6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu6"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}
layer {
  name: "drop6"
  type: "Dropout"
  bottom: "fc6"
  top: "fc6"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc7"
  type: "InnerProduct"
  bottom: "fc6"
  top: "fc7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 2
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "accuracy"
  type: "Accuracy"
  bottom: "fc7"
  bottom: "y"
  top: "accuracy"
  include {
    phase: TEST
  }
}
layer {
  name: "accuracy"
  type: "Accuracy"
  bottom: "fc7"
  bottom: "y"
  top: "accuracy"
  include {
    phase: TRAIN
  }
}
layer {
 name: "loss"
 type: "SoftmaxWithLoss"
 bottom: "fc7"
 bottom: "y"
 top: "loss"
}

• train_val_deploy.prototxt

name: "RetinaNet"
input: "data"
input_shape: {
  dim: 1
  dim: 3
  dim: 25
  dim: 25
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "conv1"
  top: "conv2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "conv2"
  top: "conv3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "conv4"
  type: "Convolution"
  bottom: "conv3"
  top: "conv4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4"
  type: "ReLU"
  bottom: "conv4"
  top: "conv4"
}
layer {
  name: "conv4_1"
  type: "Convolution"
  bottom: "conv4"
  top: "conv4_1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_1"
  type: "ReLU"
  bottom: "conv4_1"
  top: "conv4_1"
}
layer {
  name: "conv4_2"
  type: "Convolution"
  bottom: "conv4_1"
  top: "conv4_2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu4_2"
  type: "ReLU"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "fc5"
  type: "InnerProduct"
  bottom: "conv4_2"
  top: "fc5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu5"
  type: "ReLU"
  bottom: "fc5"
  top: "fc5"
}
layer {
  name: "drop5"
  type: "Dropout"
  bottom: "fc5"
  top: "fc5"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc6"
  type: "InnerProduct"
  bottom: "fc5"
  top: "fc6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu6"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}
layer {
  name: "drop6"
  type: "Dropout"
  bottom: "fc6"
  top: "fc6"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc7"
  type: "InnerProduct"
  bottom: "fc6"
  top: "fc7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 2
    weight_filler {
      type: "xavier"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
 name: "result"
 type: "Softmax"
 bottom: "fc7"
 top: "result"
}

图片数据：

序号	名称	图片大小	数量(张)	标签文件
1	img_seg_train	25*25	20520	train.txt
2	img_seg_test	25*25	5016	test.txt
3	img_seg_veri	25*25	1368	veri.txt

将图片数据转换成lmdb格式数据：

• 创建一个bat文件： lmdb.bat

C:\Caffe\caffe-master\Build\x64\Release\convert_imageset.exe --shuffle --resize_width=25 --resize_height=25 img_seg_train\ train.txt train_lmdb --backend="lmdb"
C:\Caffe\caffe-master\Build\x64\Release\convert_imageset.exe --shuffle --resize_width=25 --resize_height=25 img_seg_test\ test.txt test_lmdb --backend="lmdb"
Pause

Finetune训练：

• 创建一个bat文件，进行训练run.bat

C:\Caffe\caffe-master\Build\x64\Release\caffe.exe train -solver solver.prototxt -weights retina_20171017_21_train_iter_250000.caffemodel
Pause

其中参数：-weights retina_20171017_21_train_iter_250000.caffemodel 就代表进行finetune

试验结果

• 识别效果：

  • 标签图片

    
    
    

    标签图
    

    标签图

  • 结果图片

    
    
    

    结果图
    
    
    

    结果图

• 结果分析
  1、图片分辨率太低，背景太复杂，导致特征提取不明显

  
  
  

  25*25 像素级图片

• loss曲线

  • 未画出

更新

2017/11/01

本次更新主要内容：

• 对数据切割的大小进行调整，调整为21*21像素大小
• 考虑到原来数据集较小，通过旋转变换等，将原来图片数据集扩大了三倍，观察是否会有更好的结果，如果结果并不比之前较好，说明迁移学习的方法有待改善

将调整后的数据制作成lmdb文件，重新进行训练，漫长的等待...

更新

2017/11/04

本次更新主要内容：

• 考虑到上次识别效果特别差，猜想是网络结构的原因，故这次试验使用不同的网络架构——LeNet
• 本次试验主要参数：

net: "examples/mnist/lenet_train_test.prototxt"
test_iter: 100
test_interval: 500
base_lr: 0.01
momentum: 0.9
weight_decay: 0.0005
lr_policy: "inv"
gamma: 0.0001
power: 0.75
display: 100
max_iter: 10000
snapshot: 5000
snapshot_prefix: "examples/mnist/lenet"
solver_mode: GPU

试验效果：

• 识别图片&标签图片：

识别结果

标签图片

总体来看是被效果还是不错的，准确率为：0.907308

• 曲线图：

曲线图

总结：

• 视网膜血管图片和道路裂缝图片特征相差太大，不能使用视网膜血管中的网络模型
• 道路裂纹图片中背景复杂，干扰很大，导致在训练时提取特征的困难