基于PyCaffe实现网络可视化

前言

在进行深度学习网络调试参数的时候，往往是将整个训练过程完了之后查看训练的结果。但是除了这种办法之外还可以通过网络可视化来帮助进行参数调试，这里封装了一个小小的类，提供网络显示的接口，传入网络的名称就好了

1. 类实现

# -*-coding=utf-8-*-

import numpy as np
import matplotlib.pyplot as plt
import sys
sys.path.append('/home/sucker/Desktop/caffe-1.0/python')
import caffe

class Net_Show:
    def __init__(self, net):
        self.caffe_net = net
        self.layer_dic = {}

    # layer info(name&shape)
    def ShowLayer_Info(self):
        layer_name = [key for key, value in self.caffe_net.params.items()]
        layer_param = [value[0].data.shape for key, value in self.caffe_net.params.items()]
        for i in np.arange(0, len(layer_name)):
            self.layer_dic[layer_name[i]] = layer_param[i]

        for name in self.layer_dic:
            print name, self.layer_dic[name]

    # layer data show
    def show_data(self, data, name, 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.figure()
        plt.imshow(data, cmap='gray')
        plt.axis('off')
        plt.title(name)
        plt.rcParams['figure.figsize'] = (8, 8)
        plt.rcParams['image.interpolation'] = 'nearest'
        plt.rcParams['image.cmap'] = 'gray'
        plt.show()

	# show layer data
    def ShowBlob_Data(self, layer_name):
        data = self.caffe_net.blobs[layer_name].data[0]
        print_str = layer_name + ' data size is : ' + str(data.shape)
        print print_str

        self.show_data(data, layer_name+' data')

	# show layer params
    def ShowBlob_Param(self, layer_name):
        shape_size = np.array(self.layer_dic[layer_name])
        data = self.caffe_net.params[layer_name][0].data
        print_str = layer_name + ' param size is : ' + str(data.shape)
        print print_str
        # data = data.transpose(0, 2, 3, 1)
        data = data.reshape(shape_size[0]*shape_size[1], shape_size[2], shape_size[3])

        self.show_data(data, layer_name+' param')

2. 使用示例

caffe.set_mode_gpu()
solver = caffe.SGDSolver('solver.prototxt')
solver.step(10000)

m_show.caffe_net = solver.net
m_show.ShowBlob_Data('Convolution2')
m_show.ShowBlob_Param('Convolution1')