tf 使用 mat文件 构建vgg19 网络并进行分类预测

原文链接: tf 使用 mat文件 构建vgg19 网络并进行分类预测

上一篇: tf scipy misc 模块 tf image 模块 图片 padding 和 剪裁

下一篇: numpy top N 最大n个数

mat 文件下载

http://www.vlfeat.org/matconvnet/models/imagenet-vgg-verydeep-19.mat

参考

https://zhuanlan.zhihu.com/p/28897952

测试图片

分类结果

tiger cat :  0.9404851198196411
tabby, tabby cat :  0.03268072009086609
Egyptian cat :  0.02023828774690628

Siberian husky :  0.5219259262084961
Eskimo dog, husky :  0.4390767216682434
malamute, malemute, Alaskan malamute :  0.03156345337629318

完整代码

['tiger cat' 'Egyptian cat' 'tabby, tabby cat'] [[282 285 281]
 [250 248 249]] [0.9641481  0.01771957 0.01626895]
['Siberian husky' 'Eskimo dog, husky'
 'malamute, malemute, Alaskan malamute'] [[282 285 281]
 [250 248 249]] [0.5628869  0.40310934 0.01746248]

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import os
import scipy.io
import scipy.misc
import json
import heapq

VGG19_MAT_PATH = r"D:\迅雷下载\imagenet-vgg-verydeep-19.mat"

with open('./imagenet_1000_labels.json') as f:
    labels = json.load(f)
    labels = np.array(labels)


def _conv_layer(input, weight, bias):
    conv = tf.nn.conv2d(input, weight, strides=[1, 1, 1, 1], padding="SAME")
    return tf.nn.bias_add(conv, bias)


def _pool_layer(input):
    return tf.nn.max_pool(input, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")


def preprocess(image, mean_pixel):
    '''简单预处理,全部图片减去平均值'''
    return image - mean_pixel


def net(in_x, data_path):
    """
    读取VGG模型参数,搭建VGG网络
    :param data_path: VGG模型文件位置
    :param input_image: 输入测试图像
    :return:
    """
    layers = (
        'conv1_1', 'conv1_2', 'pool1',
        'conv2_1', 'conv2_2', 'pool2',
        'conv3_1', 'conv3_2', 'conv3_3', 'conv3_4', 'pool3',
        'conv4_1', 'conv4_2', 'conv4_3', 'conv4_4', 'pool4',
        'conv5_1', 'conv5_2', 'conv5_3', 'conv5_4', 'pool5',
        'fc1', 'fc2', 'fc3',
        'softmax'
    )
    data = scipy.io.loadmat(data_path)
    # 数据预处理的均值
    mean = data["normalization"][0][0][0][0][0]
    net = {}
    current = in_x
    net["in_x"] = in_x  # 存储数据
    count = 0  # 计数存储
    for i in range(43):
        if str(data['layers'][0][i][0][0][0][0])[:4] == "relu":
            continue
        if str(data['layers'][0][i][0][0][0][0])[:4] == "pool":
            current = _pool_layer(current)
        elif str(data['layers'][0][i][0][0][0][0]) == "softmax":
            current = tf.nn.softmax(current)
        elif i == 37:
            shape = int(np.prod(current.get_shape()[1:]))
            current = tf.reshape(current, [-1, shape])
            kernels, bias = data['layers'][0][i][0][0][0][0]
            kernels = np.reshape(kernels, [-1, 4096])
            bias = bias.reshape(-1)
            current = tf.nn.relu(tf.add(tf.matmul(current, kernels), bias))
        elif i == 39:
            kernels, bias = data['layers'][0][i][0][0][0][0]
            kernels = np.reshape(kernels, [4096, 4096])
            bias = bias.reshape(-1)
            current = tf.nn.relu(tf.add(tf.matmul(current, kernels), bias))
        elif i == 41:
            kernels, bias = data['layers'][0][i][0][0][0][0]
            kernels = np.reshape(kernels, [4096, 1000])
            bias = bias.reshape(-1)
            current = tf.add(tf.matmul(current, kernels), bias)
        else:
            kernels, bias = data['layers'][0][i][0][0][0][0]
            # 注意VGG存储方式为[,]
            # kernels = np.transpose(kernels,[1,0,2,3])
            bias = bias.reshape(-1)  # 降低维度
            current = tf.nn.relu(_conv_layer(current, kernels, bias))
        net[layers[count]] = current  # 存储数据
        count += 1
    return net


# 返回 1 224 224 3 的数据
def read_image(path):
    image = scipy.misc.imread(path, mode='RGB')
    image = scipy.misc.imresize(image, [224, 224])
    image = np.expand_dims(image, 0).astype(np.float32)
    return image


def main():
    input_image = np.concatenate([
        read_image('./images/cat.jpg'),
        read_image('./images/dog.jpg'),
    ], axis=0)

    in_x = tf.placeholder(tf.float32, (None, 224, 224, 3))
    endpoints = net(in_x, VGG19_MAT_PATH)

    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        logits_val = sess.run(endpoints, {
            in_x: input_image
        })["softmax"]

        # 取前三大概率的种类,由大到小
        labels_indexs = np.argsort(logits_val, axis=1)[:, -3:][:, ::-1]
        probs = np.sort(logits_val, axis=1)[:, -3:][:, ::-1]
        for label_index, prob in zip(labels_indexs, probs):
            print(labels[label_index], labels_indexs, prob)


if __name__ == '__main__':
    main()