《深度学习——实战caffe》——caffemodel可视化

最新推荐文章于 2024-01-15 13:42:14 发布
最新推荐文章于 2024-01-15 13:42:14 发布
阅读量375 收藏
点赞数
分类专栏： caffe 深度学习
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_33144323/article/details/80627635
版权
通过前面的学习，我们已经能够正常训练各种数据了。设置好solver.prototxt后，我们可以把训练好的模型保存起来，如lenet_iter_10000.caffemodel。训练多少次就自动保存一下，这个是通过snapshot进行设置的，保存文件的路径及文件名前缀是由snapshot_prefix来设定的。这个文件里面存放的就是各层的参数，即net.params，里面没有数据(net.blobs)。顺带还生成了一个相应的solverstate文件，这个和caffemodel差不多，但它多了一些数据，如模型名称、当前迭代次数等。两者的功能不一样，训练完后保存起来的caffemodel，是在测试阶段用来分类的，而solverstate是用来恢复训练的，防止意外终止而保存的快照（有点像断点续传的感觉)。

既然我们知道了caffemodel里面保存的就是模型各层的参数，因此我们可以把这些参数提取出来，进行可视化，看一看究竟长什么样。

我们先训练cifar10数据（mnist也可以），迭代10000次，然后将训练好的 model保存起来，名称为my_iter_10000.caffemodel，然后使用jupyter notebook 来进行可视化。

import numpy as np
import matplotlib.pyplot as plt
import os,sys,caffe  #matplotlib inline
caffe_root='/caffe/'
os.chdir(caffe_root)
sys.path.insert(0,caffe_root+'python')


# 编写一个函数，用于显示各层的参数
def show_feature(data, padsize=1, padval=0):
    data -= data.min()
    data /= data.max()

    # force the number of filters to be square
    n = int(np.ceil(np.sqrt(data.shape[0])))
    padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
    data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))

    # tile the filters into an image
    data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
    data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
    plt.imshow(data)
    plt.axis('off')

if __name__ == '__main__':
    plt.rcParams['figure.figsize'] = (8, 8)
    plt.rcParams['image.interpolation'] = 'nearest'
    plt.rcParams['image.cmap'] = 'gray'
    #设置网络模型，并显示该模型中各层名称和参数的规模（注意此处是net.params, 而不是net.blobs)
    net = caffe.Net(caffe_root + 'examples/cifar10/cifar10_quick.prototxt',
                    caffe_root + 'examples/cifar10/cifar10_quick_iter_4000.caffemodel',
                    caffe.TEST)
    [(k, v[0].data.shape) for k, v in net.params.items()]
    # 第一个卷积层，参数规模为(32,3,5,5)，即32个5*5的3通道filter
    weight = net.params["conv1"][0].data
    print(weight.shape)
    show_feature(weight.transpose(0, 2, 3, 1))
    plt.show()
参数有两种类型：权值参数和偏置项。分别用params["conv1"][0] 和params["conv1"][1] 表示 。
我们只显示权值参数，因此用params["conv1"][0]

结果如下：

WARNING: Logging before InitGoogleLogging() is written to STDERR
W0608 20:29:20.503628 23539 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface
W0608 20:29:20.503654 23539 _caffe.cpp:140] Use this instead (with the named "weights" parameter):
W0608 20:29:20.503659 23539 _caffe.cpp:142] Net('/caffe/examples/cifar10/cifar10_quick.prototxt', 1, weights='/caffe/examples/cifar10/cifar10_quick_iter_4000.caffemodel')
I0608 20:29:20.505218 23539 net.cpp:51] Initializing net from parameters: 
name: "CIFAR10_quick_test"
state {
  phase: TEST
  level: 0
}
layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param {
    shape {
      dim: 1
      dim: 3
      dim: 32
      dim: 32
    }
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "pool1"
  top: "pool1"
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 64
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3"
  top: "pool3"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "ip1"
  type: "InnerProduct"
  bottom: "pool3"
  top: "ip1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 64
  }
}
layer {
  name: "ip2"
  type: "InnerProduct"
  bottom: "ip1"
  top: "ip2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 10
  }
}
layer {
  name: "prob"
  type: "Softmax"
  bottom: "ip2"
  top: "prob"
}
I0608 20:29:20.505277 23539 layer_factory.hpp:77] Creating layer data
I0608 20:29:20.505292 23539 net.cpp:84] Creating Layer data
I0608 20:29:20.505302 23539 net.cpp:380] data -> data
I0608 20:29:20.505328 23539 net.cpp:122] Setting up data
I0608 20:29:20.505342 23539 net.cpp:129] Top shape: 1 3 32 32 (3072)
I0608 20:29:20.505350 23539 net.cpp:137] Memory required for data: 12288
I0608 20:29:20.505355 23539 layer_factory.hpp:77] Creating layer conv1
I0608 20:29:20.505364 23539 net.cpp:84] Creating Layer conv1
I0608 20:29:20.505369 23539 net.cpp:406] conv1 <- data
I0608 20:29:20.505375 23539 net.cpp:380] conv1 -> conv1
I0608 20:29:20.505400 23539 net.cpp:122] Setting up conv1
I0608 20:29:20.505409 23539 net.cpp:129] Top shape: 1 32 32 32 (32768)
I0608 20:29:20.505412 23539 net.cpp:137] Memory required for data: 143360
I0608 20:29:20.505424 23539 layer_factory.hpp:77] Creating layer pool1
I0608 20:29:20.505430 23539 net.cpp:84] Creating Layer pool1
I0608 20:29:20.505434 23539 net.cpp:406] pool1 <- conv1
I0608 20:29:20.505440 23539 net.cpp:380] pool1 -> pool1
I0608 20:29:20.505450 23539 net.cpp:122] Setting up pool1
I0608 20:29:20.505455 23539 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:29:20.505460 23539 net.cpp:137] Memory required for data: 176128
I0608 20:29:20.505463 23539 layer_factory.hpp:77] Creating layer relu1
I0608 20:29:20.505468 23539 net.cpp:84] Creating Layer relu1
I0608 20:29:20.505472 23539 net.cpp:406] relu1 <- pool1
I0608 20:29:20.505477 23539 net.cpp:367] relu1 -> pool1 (in-place)
I0608 20:29:20.505483 23539 net.cpp:122] Setting up relu1
I0608 20:29:20.505489 23539 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:29:20.505492 23539 net.cpp:137] Memory required for data: 208896
I0608 20:29:20.505496 23539 layer_factory.hpp:77] Creating layer conv2
I0608 20:29:20.505503 23539 net.cpp:84] Creating Layer conv2
I0608 20:29:20.505506 23539 net.cpp:406] conv2 <- pool1
I0608 20:29:20.505512 23539 net.cpp:380] conv2 -> conv2
I0608 20:29:20.505544 23539 net.cpp:122] Setting up conv2
I0608 20:29:20.505553 23539 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:29:20.505556 23539 net.cpp:137] Memory required for data: 241664
I0608 20:29:20.505565 23539 layer_factory.hpp:77] Creating layer relu2
I0608 20:29:20.505570 23539 net.cpp:84] Creating Layer relu2
I0608 20:29:20.505574 23539 net.cpp:406] relu2 <- conv2
I0608 20:29:20.505579 23539 net.cpp:367] relu2 -> conv2 (in-place)
I0608 20:29:20.505585 23539 net.cpp:122] Setting up relu2
I0608 20:29:20.505590 23539 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:29:20.505594 23539 net.cpp:137] Memory required for data: 274432
I0608 20:29:20.505597 23539 layer_factory.hpp:77] Creating layer pool2
I0608 20:29:20.505602 23539 net.cpp:84] Creating Layer pool2
I0608 20:29:20.505606 23539 net.cpp:406] pool2 <- conv2
I0608 20:29:20.505611 23539 net.cpp:380] pool2 -> pool2
I0608 20:29:20.505619 23539 net.cpp:122] Setting up pool2
I0608 20:29:20.505625 23539 net.cpp:129] Top shape: 1 32 8 8 (2048)
I0608 20:29:20.505627 23539 net.cpp:137] Memory required for data: 282624
I0608 20:29:20.505631 23539 layer_factory.hpp:77] Creating layer conv3
I0608 20:29:20.505637 23539 net.cpp:84] Creating Layer conv3
I0608 20:29:20.505641 23539 net.cpp:406] conv3 <- pool2
I0608 20:29:20.505646 23539 net.cpp:380] conv3 -> conv3
I0608 20:29:20.505695 23539 net.cpp:122] Setting up conv3
I0608 20:29:20.505703 23539 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:29:20.505707 23539 net.cpp:137] Memory required for data: 299008
I0608 20:29:20.505715 23539 layer_factory.hpp:77] Creating layer relu3
I0608 20:29:20.505722 23539 net.cpp:84] Creating Layer relu3
I0608 20:29:20.505726 23539 net.cpp:406] relu3 <- conv3
I0608 20:29:20.505731 23539 net.cpp:367] relu3 -> conv3 (in-place)
I0608 20:29:20.505736 23539 net.cpp:122] Setting up relu3
I0608 20:29:20.505741 23539 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:29:20.505745 23539 net.cpp:137] Memory required for data: 315392
I0608 20:29:20.505748 23539 layer_factory.hpp:77] Creating layer pool3
I0608 20:29:20.505753 23539 net.cpp:84] Creating Layer pool3
I0608 20:29:20.505758 23539 net.cpp:406] pool3 <- conv3
I0608 20:29:20.505762 23539 net.cpp:380] pool3 -> pool3
I0608 20:29:20.505769 23539 net.cpp:122] Setting up pool3
I0608 20:29:20.505774 23539 net.cpp:129] Top shape: 1 64 4 4 (1024)
I0608 20:29:20.505779 23539 net.cpp:137] Memory required for data: 319488
I0608 20:29:20.505781 23539 layer_factory.hpp:77] Creating layer ip1
I0608 20:29:20.505787 23539 net.cpp:84] Creating Layer ip1
I0608 20:29:20.505791 23539 net.cpp:406] ip1 <- pool3
I0608 20:29:20.505796 23539 net.cpp:380] ip1 -> ip1
I0608 20:29:20.505894 23539 net.cpp:122] Setting up ip1
I0608 20:29:20.505903 23539 net.cpp:129] Top shape: 1 64 (64)
I0608 20:29:20.505906 23539 net.cpp:137] Memory required for data: 319744
I0608 20:29:20.505913 23539 layer_factory.hpp:77] Creating layer ip2
I0608 20:29:20.505919 23539 net.cpp:84] Creating Layer ip2
I0608 20:29:20.505923 23539 net.cpp:406] ip2 <- ip1
I0608 20:29:20.505929 23539 net.cpp:380] ip2 -> ip2
I0608 20:29:20.505942 23539 net.cpp:122] Setting up ip2
I0608 20:29:20.505949 23539 net.cpp:129] Top shape: 1 10 (10)
I0608 20:29:20.505952 23539 net.cpp:137] Memory required for data: 319784
I0608 20:29:20.505960 23539 layer_factory.hpp:77] Creating layer prob
I0608 20:29:20.505967 23539 net.cpp:84] Creating Layer prob
I0608 20:29:20.505971 23539 net.cpp:406] prob <- ip2
I0608 20:29:20.505976 23539 net.cpp:380] prob -> prob
I0608 20:29:20.505985 23539 net.cpp:122] Setting up prob
I0608 20:29:20.505990 23539 net.cpp:129] Top shape: 1 10 (10)
I0608 20:29:20.505993 23539 net.cpp:137] Memory required for data: 319824
I0608 20:29:20.505997 23539 net.cpp:200] prob does not need backward computation.
I0608 20:29:20.506001 23539 net.cpp:200] ip2 does not need backward computation.
I0608 20:29:20.506005 23539 net.cpp:200] ip1 does not need backward computation.
I0608 20:29:20.506008 23539 net.cpp:200] pool3 does not need backward computation.
I0608 20:29:20.506012 23539 net.cpp:200] relu3 does not need backward computation.
I0608 20:29:20.506016 23539 net.cpp:200] conv3 does not need backward computation.
I0608 20:29:20.506021 23539 net.cpp:200] pool2 does not need backward computation.
I0608 20:29:20.506024 23539 net.cpp:200] relu2 does not need backward computation.
I0608 20:29:20.506028 23539 net.cpp:200] conv2 does not need backward computation.
I0608 20:29:20.506031 23539 net.cpp:200] relu1 does not need backward computation.
I0608 20:29:20.506036 23539 net.cpp:200] pool1 does not need backward computation.
I0608 20:29:20.506039 23539 net.cpp:200] conv1 does not need backward computation.
I0608 20:29:20.506043 23539 net.cpp:200] data does not need backward computation.
I0608 20:29:20.506047 23539 net.cpp:242] This network produces output prob
I0608 20:29:20.506058 23539 net.cpp:255] Network initialization done.
I0608 20:29:20.507083 23539 net.cpp:744] Ignoring source layer cifar
I0608 20:29:20.507220 23539 net.cpp:744] Ignoring source layer loss
(32, 3, 5, 5)

import numpy as np
import matplotlib.pyplot as plt
import os,sys,caffe  #matplotlib inline
caffe_root='/caffe/'
os.chdir(caffe_root)
sys.path.insert(0,caffe_root+'python')


# 编写一个函数，用于显示各层的参数
def show_feature(data, padsize=1, padval=0):
    data -= data.min()
    data /= data.max()

    # force the number of filters to be square
    n = int(np.ceil(np.sqrt(data.shape[0])))
    padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
    data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))

    # tile the filters into an image
    data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
    data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
    plt.imshow(data)
    plt.axis('off')

if __name__ == '__main__':
    plt.rcParams['figure.figsize'] = (8, 8)
    plt.rcParams['image.interpolation'] = 'nearest'
    plt.rcParams['image.cmap'] = 'gray'
    #设置网络模型，并显示该模型中各层名称和参数的规模（注意此处是net.params, 而不是net.blobs)
    net = caffe.Net(caffe_root + 'examples/cifar10/cifar10_quick.prototxt',
                    caffe_root + 'examples/cifar10/cifar10_quick_iter_4000.caffemodel',
                    caffe.TEST)
    [(k, v[0].data.shape) for k, v in net.params.items()]
    # 第一个卷积层，参数规模为(32,3,5,5)，即32个5*5的3通道filter
    #weight = net.params["conv1"][0].data
    #print(weight.shape)
    #show_feature(weight.transpose(0, 2, 3, 1))

    # 第二个卷积层的权值参数，共有32*32个filter,每个filter大小为5*5
    weight = net.params["conv2"][0].data
    print(weight.shape)
    show_feature(weight.reshape(32 ** 2, 5, 5))

    plt.show()

结果如下：

WARNING: Logging before InitGoogleLogging() is written to STDERR
W0608 20:52:40.430717 24894 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface
W0608 20:52:40.430750 24894 _caffe.cpp:140] Use this instead (with the named "weights" parameter):
W0608 20:52:40.430754 24894 _caffe.cpp:142] Net('/caffe/examples/cifar10/cifar10_quick.prototxt', 1, weights='/caffe/examples/cifar10/cifar10_quick_iter_4000.caffemodel')
I0608 20:52:40.432391 24894 net.cpp:51] Initializing net from parameters: 
name: "CIFAR10_quick_test"
state {
  phase: TEST
  level: 0
}
layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param {
    shape {
      dim: 1
      dim: 3
      dim: 32
      dim: 32
    }
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "pool1"
  top: "pool1"
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 64
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3"
  top: "pool3"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "ip1"
  type: "InnerProduct"
  bottom: "pool3"
  top: "ip1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 64
  }
}
layer {
  name: "ip2"
  type: "InnerProduct"
  bottom: "ip1"
  top: "ip2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 10
  }
}
layer {
  name: "prob"
  type: "Softmax"
  bottom: "ip2"
  top: "prob"
}
I0608 20:52:40.432447 24894 layer_factory.hpp:77] Creating layer data
I0608 20:52:40.432457 24894 net.cpp:84] Creating Layer data
I0608 20:52:40.432463 24894 net.cpp:380] data -> data
I0608 20:52:40.432482 24894 net.cpp:122] Setting up data
I0608 20:52:40.432490 24894 net.cpp:129] Top shape: 1 3 32 32 (3072)
I0608 20:52:40.432493 24894 net.cpp:137] Memory required for data: 12288
I0608 20:52:40.432497 24894 layer_factory.hpp:77] Creating layer conv1
I0608 20:52:40.432505 24894 net.cpp:84] Creating Layer conv1
I0608 20:52:40.432510 24894 net.cpp:406] conv1 <- data
I0608 20:52:40.432516 24894 net.cpp:380] conv1 -> conv1
I0608 20:52:40.432541 24894 net.cpp:122] Setting up conv1
I0608 20:52:40.432548 24894 net.cpp:129] Top shape: 1 32 32 32 (32768)
I0608 20:52:40.432552 24894 net.cpp:137] Memory required for data: 143360
I0608 20:52:40.432564 24894 layer_factory.hpp:77] Creating layer pool1
I0608 20:52:40.432570 24894 net.cpp:84] Creating Layer pool1
I0608 20:52:40.432574 24894 net.cpp:406] pool1 <- conv1
I0608 20:52:40.432579 24894 net.cpp:380] pool1 -> pool1
I0608 20:52:40.432590 24894 net.cpp:122] Setting up pool1
I0608 20:52:40.432595 24894 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:52:40.432598 24894 net.cpp:137] Memory required for data: 176128
I0608 20:52:40.432602 24894 layer_factory.hpp:77] Creating layer relu1
I0608 20:52:40.432607 24894 net.cpp:84] Creating Layer relu1
I0608 20:52:40.432611 24894 net.cpp:406] relu1 <- pool1
I0608 20:52:40.432615 24894 net.cpp:367] relu1 -> pool1 (in-place)
I0608 20:52:40.432621 24894 net.cpp:122] Setting up relu1
I0608 20:52:40.432626 24894 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:52:40.432631 24894 net.cpp:137] Memory required for data: 208896
I0608 20:52:40.432633 24894 layer_factory.hpp:77] Creating layer conv2
I0608 20:52:40.432641 24894 net.cpp:84] Creating Layer conv2
I0608 20:52:40.432643 24894 net.cpp:406] conv2 <- pool1
I0608 20:52:40.432649 24894 net.cpp:380] conv2 -> conv2
I0608 20:52:40.432680 24894 net.cpp:122] Setting up conv2
I0608 20:52:40.432690 24894 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:52:40.432693 24894 net.cpp:137] Memory required for data: 241664
I0608 20:52:40.432701 24894 layer_factory.hpp:77] Creating layer relu2
I0608 20:52:40.432708 24894 net.cpp:84] Creating Layer relu2
I0608 20:52:40.432711 24894 net.cpp:406] relu2 <- conv2
I0608 20:52:40.432716 24894 net.cpp:367] relu2 -> conv2 (in-place)
I0608 20:52:40.432723 24894 net.cpp:122] Setting up relu2
I0608 20:52:40.432729 24894 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:52:40.432732 24894 net.cpp:137] Memory required for data: 274432
I0608 20:52:40.432735 24894 layer_factory.hpp:77] Creating layer pool2
I0608 20:52:40.432740 24894 net.cpp:84] Creating Layer pool2
I0608 20:52:40.432744 24894 net.cpp:406] pool2 <- conv2
I0608 20:52:40.432749 24894 net.cpp:380] pool2 -> pool2
I0608 20:52:40.432756 24894 net.cpp:122] Setting up pool2
I0608 20:52:40.432761 24894 net.cpp:129] Top shape: 1 32 8 8 (2048)
I0608 20:52:40.432765 24894 net.cpp:137] Memory required for data: 282624
I0608 20:52:40.432768 24894 layer_factory.hpp:77] Creating layer conv3
I0608 20:52:40.432775 24894 net.cpp:84] Creating Layer conv3
I0608 20:52:40.432778 24894 net.cpp:406] conv3 <- pool2
I0608 20:52:40.432783 24894 net.cpp:380] conv3 -> conv3
I0608 20:52:40.432834 24894 net.cpp:122] Setting up conv3
I0608 20:52:40.432842 24894 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:52:40.432845 24894 net.cpp:137] Memory required for data: 299008
I0608 20:52:40.432857 24894 layer_factory.hpp:77] Creating layer relu3
I0608 20:52:40.432866 24894 net.cpp:84] Creating Layer relu3
I0608 20:52:40.432873 24894 net.cpp:406] relu3 <- conv3
I0608 20:52:40.432880 24894 net.cpp:367] relu3 -> conv3 (in-place)
I0608 20:52:40.432889 24894 net.cpp:122] Setting up relu3
I0608 20:52:40.432895 24894 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:52:40.432898 24894 net.cpp:137] Memory required for data: 315392
I0608 20:52:40.432902 24894 layer_factory.hpp:77] Creating layer pool3
I0608 20:52:40.432907 24894 net.cpp:84] Creating Layer pool3
I0608 20:52:40.432910 24894 net.cpp:406] pool3 <- conv3
I0608 20:52:40.432915 24894 net.cpp:380] pool3 -> pool3
I0608 20:52:40.432922 24894 net.cpp:122] Setting up pool3
I0608 20:52:40.432927 24894 net.cpp:129] Top shape: 1 64 4 4 (1024)
I0608 20:52:40.432931 24894 net.cpp:137] Memory required for data: 319488
I0608 20:52:40.432934 24894 layer_factory.hpp:77] Creating layer ip1
I0608 20:52:40.432940 24894 net.cpp:84] Creating Layer ip1
I0608 20:52:40.432945 24894 net.cpp:406] ip1 <- pool3
I0608 20:52:40.432950 24894 net.cpp:380] ip1 -> ip1
I0608 20:52:40.433044 24894 net.cpp:122] Setting up ip1
I0608 20:52:40.433053 24894 net.cpp:129] Top shape: 1 64 (64)
I0608 20:52:40.433056 24894 net.cpp:137] Memory required for data: 319744
I0608 20:52:40.433063 24894 layer_factory.hpp:77] Creating layer ip2
I0608 20:52:40.433070 24894 net.cpp:84] Creating Layer ip2
I0608 20:52:40.433074 24894 net.cpp:406] ip2 <- ip1
I0608 20:52:40.433080 24894 net.cpp:380] ip2 -> ip2
I0608 20:52:40.433094 24894 net.cpp:122] Setting up ip2
I0608 20:52:40.433099 24894 net.cpp:129] Top shape: 1 10 (10)
I0608 20:52:40.433102 24894 net.cpp:137] Memory required for data: 319784
I0608 20:52:40.433110 24894 layer_factory.hpp:77] Creating layer prob
I0608 20:52:40.433116 24894 net.cpp:84] Creating Layer prob
I0608 20:52:40.433120 24894 net.cpp:406] prob <- ip2
I0608 20:52:40.433128 24894 net.cpp:380] prob -> prob
I0608 20:52:40.433137 24894 net.cpp:122] Setting up prob
I0608 20:52:40.433142 24894 net.cpp:129] Top shape: 1 10 (10)
I0608 20:52:40.433145 24894 net.cpp:137] Memory required for data: 319824
I0608 20:52:40.433149 24894 net.cpp:200] prob does not need backward computation.
I0608 20:52:40.433152 24894 net.cpp:200] ip2 does not need backward computation.
I0608 20:52:40.433156 24894 net.cpp:200] ip1 does not need backward computation.
I0608 20:52:40.433161 24894 net.cpp:200] pool3 does not need backward computation.
I0608 20:52:40.433163 24894 net.cpp:200] relu3 does not need backward computation.
I0608 20:52:40.433167 24894 net.cpp:200] conv3 does not need backward computation.
I0608 20:52:40.433171 24894 net.cpp:200] pool2 does not need backward computation.
I0608 20:52:40.433174 24894 net.cpp:200] relu2 does not need backward computation.
I0608 20:52:40.433177 24894 net.cpp:200] conv2 does not need backward computation.
I0608 20:52:40.433182 24894 net.cpp:200] relu1 does not need backward computation.
(32, 32, 5, 5)
I0608 20:52:40.433184 24894 net.cpp:200] pool1 does not need backward computation.
I0608 20:52:40.433188 24894 net.cpp:200] conv1 does not need backward computation.
I0608 20:52:40.433192 24894 net.cpp:200] data does not need backward computation.
I0608 20:52:40.433195 24894 net.cpp:242] This network produces output prob
I0608 20:52:40.433204 24894 net.cpp:255] Network initialization done.
I0608 20:52:40.434226 24894 net.cpp:744] Ignoring source layer cifar
I0608 20:52:40.434352 24894 net.cpp:744] Ignoring source layer loss

import numpy as np
import matplotlib.pyplot as plt
import os,sys,caffe  #matplotlib inline
caffe_root='/caffe/'
os.chdir(caffe_root)
sys.path.insert(0,caffe_root+'python')


# 编写一个函数，用于显示各层的参数
def show_feature(data, padsize=1, padval=0):
    data -= data.min()
    data /= data.max()

    # force the number of filters to be square
    n = int(np.ceil(np.sqrt(data.shape[0])))
    padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
    data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))

    # tile the filters into an image
    data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
    data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
    plt.imshow(data)
    plt.axis('off')

if __name__ == '__main__':
    plt.rcParams['figure.figsize'] = (8, 8)
    plt.rcParams['image.interpolation'] = 'nearest'
    plt.rcParams['image.cmap'] = 'gray'
    #设置网络模型，并显示该模型中各层名称和参数的规模（注意此处是net.params, 而不是net.blobs)
    net = caffe.Net(caffe_root + 'examples/cifar10/cifar10_quick.prototxt',
                    caffe_root + 'examples/cifar10/cifar10_quick_iter_4000.caffemodel',
                    caffe.TEST)
    [(k, v[0].data.shape) for k, v in net.params.items()]
    # 第一个卷积层，参数规模为(32,3,5,5)，即32个5*5的3通道filter
    #weight = net.params["conv1"][0].data
    #print(weight.shape)
    #show_feature(weight.transpose(0, 2, 3, 1))

    # 第二个卷积层的权值参数，共有32*32个filter,每个filter大小为5*5
    #weight = net.params["conv2"][0].data
    #print(weight.shape)
    #show_feature(weight.reshape(32 ** 2, 5, 5))

    # 第三个卷积层的权值，共有64*32个filter,每个filter大小为5*5，取其前1024个进行可视化
    weight = net.params["conv3"][0].data
    print(weight.shape)
    show_feature(weight.reshape(64 * 32, 5, 5)[:1024])

    plt.show()

结果如下：

WARNING: Logging before InitGoogleLogging() is written to STDERR
W0608 20:57:26.513041 25238 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface
W0608 20:57:26.513070 25238 _caffe.cpp:140] Use this instead (with the named "weights" parameter):
W0608 20:57:26.513073 25238 _caffe.cpp:142] Net('/caffe/examples/cifar10/cifar10_quick.prototxt', 1, weights='/caffe/examples/cifar10/cifar10_quick_iter_4000.caffemodel')
I0608 20:57:26.514796 25238 net.cpp:51] Initializing net from parameters: 
name: "CIFAR10_quick_test"
state {
  phase: TEST
  level: 0
}
layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param {
    shape {
      dim: 1
      dim: 3
      dim: 32
      dim: 32
    }
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "pool1"
  top: "pool1"
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 32
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "pool2"
  top: "conv3"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 64
    pad: 2
    kernel_size: 5
    stride: 1
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv3"
  top: "pool3"
  pooling_param {
    pool: AVE
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "ip1"
  type: "InnerProduct"
  bottom: "pool3"
  top: "ip1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 64
  }
}
layer {
  name: "ip2"
  type: "InnerProduct"
  bottom: "ip1"
  top: "ip2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 10
  }
}
layer {
  name: "prob"
  type: "Softmax"
  bottom: "ip2"
  top: "prob"
}
I0608 20:57:26.514855 25238 layer_factory.hpp:77] Creating layer data
I0608 20:57:26.514871 25238 net.cpp:84] Creating Layer data
I0608 20:57:26.514881 25238 net.cpp:380] data -> data
I0608 20:57:26.514909 25238 net.cpp:122] Setting up data
I0608 20:57:26.514924 25238 net.cpp:129] Top shape: 1 3 32 32 (3072)
I0608 20:57:26.514943 25238 net.cpp:137] Memory required for data: 12288
I0608 20:57:26.514950 25238 layer_factory.hpp:77] Creating layer conv1
I0608 20:57:26.514964 25238 net.cpp:84] Creating Layer conv1
I0608 20:57:26.514972 25238 net.cpp:406] conv1 <- data
I0608 20:57:26.514983 25238 net.cpp:380] conv1 -> conv1
I0608 20:57:26.515027 25238 net.cpp:122] Setting up conv1
I0608 20:57:26.515039 25238 net.cpp:129] Top shape: 1 32 32 32 (32768)
I0608 20:57:26.515048 25238 net.cpp:137] Memory required for data: 143360
I0608 20:57:26.515064 25238 layer_factory.hpp:77] Creating layer pool1
I0608 20:57:26.515076 25238 net.cpp:84] Creating Layer pool1
I0608 20:57:26.515084 25238 net.cpp:406] pool1 <- conv1
I0608 20:57:26.515091 25238 net.cpp:380] pool1 -> pool1
I0608 20:57:26.515106 25238 net.cpp:122] Setting up pool1
I0608 20:57:26.515118 25238 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:57:26.515125 25238 net.cpp:137] Memory required for data: 176128
I0608 20:57:26.515132 25238 layer_factory.hpp:77] Creating layer relu1
I0608 20:57:26.515141 25238 net.cpp:84] Creating Layer relu1
I0608 20:57:26.515147 25238 net.cpp:406] relu1 <- pool1
I0608 20:57:26.515153 25238 net.cpp:367] relu1 -> pool1 (in-place)
I0608 20:57:26.515159 25238 net.cpp:122] Setting up relu1
I0608 20:57:26.515166 25238 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:57:26.515168 25238 net.cpp:137] Memory required for data: 208896
I0608 20:57:26.515172 25238 layer_factory.hpp:77] Creating layer conv2
I0608 20:57:26.515179 25238 net.cpp:84] Creating Layer conv2
I0608 20:57:26.515183 25238 net.cpp:406] conv2 <- pool1
I0608 20:57:26.515189 25238 net.cpp:380] conv2 -> conv2
I0608 20:57:26.515231 25238 net.cpp:122] Setting up conv2
I0608 20:57:26.515244 25238 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:57:26.515250 25238 net.cpp:137] Memory required for data: 241664
I0608 20:57:26.515264 25238 layer_factory.hpp:77] Creating layer relu2
I0608 20:57:26.515274 25238 net.cpp:84] Creating Layer relu2
I0608 20:57:26.515280 25238 net.cpp:406] relu2 <- conv2
I0608 20:57:26.515286 25238 net.cpp:367] relu2 -> conv2 (in-place)
I0608 20:57:26.515296 25238 net.cpp:122] Setting up relu2
I0608 20:57:26.515307 25238 net.cpp:129] Top shape: 1 32 16 16 (8192)
I0608 20:57:26.515313 25238 net.cpp:137] Memory required for data: 274432
I0608 20:57:26.515319 25238 layer_factory.hpp:77] Creating layer pool2
I0608 20:57:26.515327 25238 net.cpp:84] Creating Layer pool2
I0608 20:57:26.515336 25238 net.cpp:406] pool2 <- conv2
I0608 20:57:26.515344 25238 net.cpp:380] pool2 -> pool2
I0608 20:57:26.515357 25238 net.cpp:122] Setting up pool2
I0608 20:57:26.515369 25238 net.cpp:129] Top shape: 1 32 8 8 (2048)
I0608 20:57:26.515377 25238 net.cpp:137] Memory required for data: 282624
I0608 20:57:26.515383 25238 layer_factory.hpp:77] Creating layer conv3
I0608 20:57:26.515395 25238 net.cpp:84] Creating Layer conv3
I0608 20:57:26.515403 25238 net.cpp:406] conv3 <- pool2
I0608 20:57:26.515414 25238 net.cpp:380] conv3 -> conv3
I0608 20:57:26.515506 25238 net.cpp:122] Setting up conv3
I0608 20:57:26.515519 25238 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:57:26.515527 25238 net.cpp:137] Memory required for data: 299008
I0608 20:57:26.515542 25238 layer_factory.hpp:77] Creating layer relu3
I0608 20:57:26.515554 25238 net.cpp:84] Creating Layer relu3
I0608 20:57:26.515563 25238 net.cpp:406] relu3 <- conv3
I0608 20:57:26.515573 25238 net.cpp:367] relu3 -> conv3 (in-place)
I0608 20:57:26.515583 25238 net.cpp:122] Setting up relu3
I0608 20:57:26.515592 25238 net.cpp:129] Top shape: 1 64 8 8 (4096)
I0608 20:57:26.515599 25238 net.cpp:137] Memory required for data: 315392
I0608 20:57:26.515605 25238 layer_factory.hpp:77] Creating layer pool3
I0608 20:57:26.515615 25238 net.cpp:84] Creating Layer pool3
I0608 20:57:26.515622 25238 net.cpp:406] pool3 <- conv3
I0608 20:57:26.515632 25238 net.cpp:380] pool3 -> pool3
I0608 20:57:26.515645 25238 net.cpp:122] Setting up pool3
I0608 20:57:26.515655 25238 net.cpp:129] Top shape: 1 64 4 4 (1024)
I0608 20:57:26.515662 25238 net.cpp:137] Memory required for data: 319488
I0608 20:57:26.515669 25238 layer_factory.hpp:77] Creating layer ip1
I0608 20:57:26.515681 25238 net.cpp:84] Creating Layer ip1
I0608 20:57:26.515687 25238 net.cpp:406] ip1 <- pool3
I0608 20:57:26.515698 25238 net.cpp:380] ip1 -> ip1
I0608 20:57:26.515851 25238 net.cpp:122] Setting up ip1
I0608 20:57:26.515863 25238 net.cpp:129] Top shape: 1 64 (64)
I0608 20:57:26.515869 25238 net.cpp:137] Memory required for data: 319744
I0608 20:57:26.515882 25238 layer_factory.hpp:77] Creating layer ip2
I0608 20:57:26.515892 25238 net.cpp:84] Creating Layer ip2
I0608 20:57:26.515900 25238 net.cpp:406] ip2 <- ip1
I0608 20:57:26.515911 25238 net.cpp:380] ip2 -> ip2
I0608 20:57:26.515933 25238 net.cpp:122] Setting up ip2
I0608 20:57:26.515944 25238 net.cpp:129] Top shape: 1 10 (10)
I0608 20:57:26.515951 25238 net.cpp:137] Memory required for data: 319784
I0608 20:57:26.515966 25238 layer_factory.hpp:77] Creating layer prob
I0608 20:57:26.515980 25238 net.cpp:84] Creating Layer prob
I0608 20:57:26.515987 25238 net.cpp:406] prob <- ip2
I0608 20:57:26.515997 25238 net.cpp:380] prob -> prob
I0608 20:57:26.516012 25238 net.cpp:122] Setting up prob
I0608 20:57:26.516021 25238 net.cpp:129] Top shape: 1 10 (10)
I0608 20:57:26.516028 25238 net.cpp:137] Memory required for data: 319824
I0608 20:57:26.516036 25238 net.cpp:200] prob does not need backward computation.
I0608 20:57:26.516042 25238 net.cpp:200] ip2 does not need backward computation.
I0608 20:57:26.516050 25238 net.cpp:200] ip1 does not need backward computation.
I0608 20:57:26.516057 25238 net.cpp:200] pool3 does not need backward computation.
I0608 20:57:26.516064 25238 net.cpp:200] relu3 does not need backward computation.
I0608 20:57:26.516072 25238 net.cpp:200] conv3 does not need backward computation.
I0608 20:57:26.516079 25238 net.cpp:200] pool2 does not need backward computation.
I0608 20:57:26.516086 25238 net.cpp:200] relu2 does not need backward computation.
I0608 20:57:26.516093 25238 net.cpp:200] conv2 does not need backward computation.
I0608 20:57:26.516100 25238 net.cpp:200] relu1 does not need backward computation.
I0608 20:57:26.516108 25238 net.cpp:200] pool1 does not need backward computation.
I0608 20:57:26.516115 25238 net.cpp:200] conv1 does not need backward computation.
I0608 20:57:26.516124 25238 net.cpp:200] data does not need backward computation.
I0608 20:57:26.516129 25238 net.cpp:242] This network produces output prob
I0608 20:57:26.516146 25238 net.cpp:255] Network initialization done.
I0608 20:57:26.517181 25238 net.cpp:744] Ignoring source layer cifar
I0608 20:57:26.517324 25238 net.cpp:744] Ignoring source layer loss
(64, 32, 5, 5)


频率52HZ
关注
  • 0
    点赞
  • 踩
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
专栏目录
caffe 跑模型报错 Error parsing text-format caffe.NetParameter: 3707:25: Message type "caffe.LayerParamet
01-29 7513
WARNING: Logging before InitGoogleLogging() is written to STDERR W0129 17:35:40.717015 14357 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface W0129 17:35:40.717033 14357 _caf
Caffe框架的机器学习——安装与问题锦集
qq_42450724的博客
06-17 1368
Ubuntu16.04   python   Faster R-CNN安装教程：一、Caffe安装教程：Ubuntu16.04(CPU)   https://blog.csdn.net/u010193446/article/details/53259294二、Faster R-CNN 的Caffe实现https://www.2cto.com/kf/201706/646213.html三、Faste...
WARNING: Logging before InitGoogleLogging() is written to STDERR 解决办法
我在玩剑三的博客
11-12 1万+
这里写自定义目录标题欢迎使用Markdown编辑器新的改变功能快捷键合理的创建标题，有助于目录的生成如何改变文本的样式插入链接与图片如何插入一段漂亮的代码片生成一个适合你的列表创建一个表格设定内容居中、居左、居右SmartyPants创建一个自定义列表如何创建一个注脚注释也是必不可少的KaTeX数学公式新的甘特图功能，丰富你的文章UML 图表FLowchart流程图导出与导入导出导入 欢迎使用Markdown编辑器 你好！ 这是你第一次使用 Markdown编辑器 所展示的欢迎页。如果你想学习如何使用Mar
使用mindspore产生随机数据集时shuffle出现bug
weixin_45666880的博客
06-20 569
问题描述：【操作步骤&问题现象】产生随机数据集时shuffle出现bug【截图信息】 【日志信息】（可选，上传日志内容或者附件）报错信息如下所示：WARNING: Logging before InitGoogleLogging() is written to STDERR [ERROR] MD(12708,9,?):2022-4-22 16:52:18 [mindspore\ccsrc\minddata\dataset\util\task_manager.cc:217] InterruptMaster
"caffe.LayerParameter" has no field named
confuciust的专栏
06-05 5148
详细报错内容: WARNING: Logging before InitGoogleLogging() is written to STDERR W0605 14:58:08.718324 25393 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface W0605 14:58:08.718341 2...
深度学习_21天实战Caffe+Caffe之经典模型详解与实战 (pdf文件 带书签)
04-11
《深度学习：21天实战Caffe+Caffe之经典模型详解与实战》是一本专注于Caffe框架的深度学习实践指南，旨在帮助读者快速掌握Caffe并应用到实际项目中。Caffe是目前广泛使用的深度学习库之一，以其高效、易用和模块化的...
深度学习caffe实战
12-22
接下来，书中会详细讲解Caffe的模型定义语言（prototxt）和模型存储格式（caffemodel），这是构建和训练模型的关键。读者将学会如何编写网络结构文件，设置超参数，并理解Caffe中的层类型，如卷积层、池化层、全连接...
深度学习21天实战Caffe完整版
01-20
书籍会详细解析Caffe的网络定义（.prototxt）、模型权重存储（.caffemodel）以及如何通过这些文件构建和训练深度学习模型。 3. **Caffe安装与配置**：为了实践，书籍会提供Caffe在不同平台（如Linux、Windows和Mac ...
视觉SLAM14讲中ch13运行程序时出现警告WARNING: Logging before InitGoogleLogging() is written to STDERR解决方案
最新发布
m0_49384824的博客
01-15 931
gflags是一个谷歌的库,可以很方便地定义一些全局变量，并且可以从命令行设置他们的值。官方文档库为https://gflags.github.io/gflags/DEFINE_XXX(name, val, txt)是一个宏，他的作用是创建一个XXX类型全局变量，该变量的值可以在程序运行时通过命令行参数进行设置。在程序中使用DEFINE_XXX宏定义的变量时，需要在变量前加上FLAGS_前缀。依次输出变量名（变量的提示信息） 变量类型 变量默认值。对于从命令行设置全局变量的值，即命令行输入格式，
caffe不能使用java开发_caffe：caffemodel无法加载的原因？
weixin_39619858的博客
02-13 185
我一年前在机器学习中的冒险开始于caffe . 我使用python接口，并使用 caffe/examples/ 目录中提供的ipython笔记本学到很多东西 .成功运行笔记本01-learning-lenet.ipynb并且能够加载我创建的caffe模型( .caffemodel 文件)后，我冒险将相同的LeNet神经网络应用到我自己的数据中进行字符识别 .所以我使用了convert_images...
caffe学习:caffemodel可视化
hukai7190的博客
01-15 1828
网络架构的形象展示在构建完网络后, 我们得到一个xx_train.prototxt文件, 该文件中记录着每层网络的输入输出及规格等参数, 在地址http://ethereon.github.io/netscope/#/editor 左侧拖入xx_train.ptototxt文件, 按enter+shift组合键即可查看网络模型图, 非常具体形象.
caffe网络模型可视化方法
ZONGXP的博客
07-11 9162
参考：https://blog.csdn.net/u012915263/article/details/60579429caffe的*.prototxt定义了神经网络的结构，为了直观的看到神经网络的连接工具，使用Netscope工具来在线显示网络结构链接：https://ethereon.github.io/netscope/#/editor注意该网需要翻墙使用打开后界面如图所示将*.protot...
Caffe学习之模型可视化
g110802008的博客
12-06 348
1、网络可视化 准备Python环境 因为caffe是使用基于Python的网络结构可视化工具，所以需要Python环境，即编译pycaffe后就可使用。 # cd ${CAFFE_ROOT} # make clean # make -j16 # make pycaffe -j16 顺便编译一下MATLAB的环境 # make matcaffe -j16 绘制网络结构图 # cd ${CAFFE
Caffe模型可视化
RTM的博客
03-01 2558
1、caffe模型可视化 下面的网站可以将caffe模型可视化，只要你输入自己的caffe网络配置文件。 网站地址：http://ethereon.github.io/netscope/quickstart.html 2、卷积神经网络可视化 这个网站是CS231n的讲师karpathy做的一个网站，里面有好多有意思的东西。其中包括CIFAR-10数据集的训练过程,可以可是化的看到每一层卷
Caffe 的可视化 （三） caffe model 的可视化
frontworld的博客
11-23 1729
Caffe 的可视化 （三） caffe model 的可视化 以 cifar10 quick net 为例子， 首先下载cifar10 data并且训练得到model: #cd to the caffe root mark@ubuntu:~$ cd caffe  #download the cifar10 data mark@ubuntu:~/caffe$ ./da
可视化caffe模型结构及在线可视化
jiandanjinxin的专栏
02-18 5771
假设Caffe的目录是$(CAFFE_ROOT) 1.编译caffe的python接口$ make pycaffe2.装各种依赖$ pip install pydot $ sudo apt-get install graphviz3.可视化模型$ cd $(CAFFE_ROOT)/python/ $ ./draw_net.py ../models/bvlc_googlenet/deploy.pro
将图像数据jpg,png等存储为npy/npz格式
热门推荐
qq_38410428的博客
10-16 1万+
前言 将目录./Data/RSD19的19种图像数据的图像，按种类存储为npy/npz格式。RSD19的子目录为，每一个子目录下都有相应种类的图像。 思路 首先遍历根目录下所有子目录和其中的文件，将内容与label一起保存在numpy中。 打乱顺序。 输出四个npy/npz文件。 于是可以写成下面两个函数。 首先是一个辅助的函数, 用来返回label的。例如文件夹内容为Airport, Beach, 最后可以范围对应关系, {‘Airport’: 0, ‘Beach’: 1}。 接下来就
CentOS7下glob库安装与测试
没有最好，只有更好！
08-07 1750
glob库 glob是google出的一个C++轻量级日志库，支持以下功能： 参数设置，以命令行参数的方式设置标志参数来控制日志记录行为 严重性分级，根据日志严重性分级记录日志 可有条件地记录日志信息 条件终止程序。丰富的条件判断宏，可预设程序终止条件 异常信号处理。程序异常情况，可自定义异常处理过程 支持debug功能 自定义日志信息 线程安全日志记录方式 系统级日志记录 ...
《深度学习：21天实战Caffe》读书笔记模板.pptx
09-14
深度学习：21天实战Caffe是一本以深度学习框架Caffe为核心，针对初学者设计的实战教程。该书注重理论与实践的结合，通过21天的学习路径，逐步引导读者掌握Caffe的配置、部署和使用方法，从而深入理解深度学习的基本...

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值