caffe mnist实战训练（5）用可视化图片解决问题

前面记录了生成lmdb文件和模型，模型的准确率也达到了0.98，现在开始用一张实际的图片来实验

先写一个label.txt文件，对应数字的标签，内容如下

0
1
2
3
4
5
6
7
8
9

再编写一个deploy.protoxtx文件，代码如下

name:"Lenet"
input:"data"
input_dim:1
input_dim:3
input_dim:28
input_dim:28
layer {
  name: "Convolution1"
  type: "Convolution"
  bottom: "data"
  top: "Convolution1"
  convolution_param {
    num_output: 20
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Pooling1"
  type: "Pooling"
  bottom: "Convolution1"
  top: "Pooling1"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "Convolution2"
  type: "Convolution"
  bottom: "Pooling1"
  top: "Convolution2"
  convolution_param {
    num_output: 50
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Pooling2"
  type: "Pooling"
  bottom: "Convolution2"
  top: "Pooling2"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "InnerProduct1"
  type: "InnerProduct"
  bottom: "Pooling2"
  top: "InnerProduct1"
  inner_product_param {
    num_output: 500
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "ReLU1"
  type: "ReLU"
  bottom: "InnerProduct1"
  top: "InnerProduct1"
}
layer {
  name: "InnerProduct2"
  type: "InnerProduct"
  bottom: "InnerProduct1"
  top: "InnerProduct2"
  inner_product_param {
    num_output: 10
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "Softmax1"
  type: "Softmax"
  bottom: "InnerProduct2"
  top: "Softmax1"
}

然后编写一个verify.py文件，注意路径就可以了

# coding=utf-8
import sys
sys.path.append('/home/xhj/caffe/python')
import caffe
import numpy as np

root = '/home/xhj/hjxu-code/matlabcode/MNIST-TEST/mnist/'  # 根目录
deploy = root + 'profile/pro/deploy.prototxt'  # deploy文件
caffe_model = root + 'model/caffe_model_mnist_iter_10000.caffemodel'  # 训练好的 caffemodel
img = root + 'test/3/00032.png'  # 随机找的一张待测图片
labels_filename = root + 'profile/labels.txt'  # 类别名称文件，将数字标签转换回类别名称
#mean_file = root + 'profile/train_mean.npy'   #加载均值文件
net = caffe.Net(deploy, caffe_model, caffe.TEST)  # 加载model和network

# 图片预处理设置
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})  # 设定图片的shape格式(1,3,28,28)
transformer.set_transpose('data', (2, 0, 1))  # 改变维度的顺序，由原始图片(28,28,3)变为(3,28,28)
#transformer.set_mean('data', np.load(mean_file).mean(1).mean(1))    # 减去均值，前面训练模型时没有减均值，这儿就不用
transformer.set_raw_scale('data', 255)  # 缩放到【0，255】之间
transformer.set_channel_swap('data', (2, 1, 0))  # 交换通道，将图片由RGB变为BGR

im = caffe.io.load_image(img)  # 加载图片
net.blobs['data'].data[...] = transformer.preprocess('data', im)  # 执行上面设置的图片预处理操作，并将图片载入到blob中

# 执行测试
out = net.forward()

labels = np.loadtxt(labels_filename, str, delimiter='\t')  # 读取类别名称文件
prob = net.blobs['prob'].data[0].flatten()  # 取出最后一层（Softmax）属于某个类别的概率值，并打印
print prob
order = prob.argsort()[-1]  # 将概率值排序，取出最大值所在的序号
print 'the class is:',labels[order]   #将该序号转换成对应的类别名称，并打印

得到结果

[ 0.  0.  0.  1.  0.  0.  0.  0.  0.  0.]
the class is: 3

Process finished with exit code 0

到此mnist手写数字识别实验成功结束了