数据集：CASIA-FaceV5 下载地址：http://biometrics.idealtest.org/dbDetailForUser.do?id=9
数据集介绍：总共2500张图片，500个人，每个人有5张图片（彩色图像），每一张图像大小为640×480
我的训练集有2100张图片，480类，测试集有220张图片
import tensorflow as tf import numpy as np import matplotlib.pyplot as plt import loaddata import _pickle as pickle import seaborn class CnnChaoCan(object): # 模型参数 learning_rate = 0.00001 training_iters =40 batch_size = 70 display_step = 5 n_features =37632 #112*112*3 n_classes = 480 n_fc1 = 384 n_fc2 = 4096 class faceCNN(object): def __init__(self,config): # 构建模型 self.config=config self.x = tf.placeholder(tf.float32, [None, 112,112,3]) self.y = tf.placeholder(tf.float32, [None, config.n_classes]) self.keep_prob=tf.placeholder(tf.float32) self.w = { 'conv1': tf.Variable(tf.truncated_normal([5, 5, 3, 64], stddev=5e-2)), 'conv2': tf.Variable(tf.truncated_normal([5, 5, 64, 64], stddev=0.1)), 'fc1': tf.Variable(tf.truncated_normal([28*28* 64, config.n_fc1], stddev=0.04)), 'fc2': tf.Variable(tf.truncated_normal([config.n_fc1, config.n_fc2], stddev=0.1)), 'fc3': tf.Variable(tf.truncated_normal([config.n_fc2, config.n_classes], stddev=1 / 192.0)) } self.b = { 'conv1': tf.Variable(tf.constant(0.0, dtype=tf.float32, shape=[64])), 'conv2': tf.Variable(tf.constant(0.1, dtype=tf.float32, shape=[64])), 'fc1': tf.Variable(tf.constant(0.1, dtype=tf.float32, shape=[config.n_fc1])), 'fc2': tf.Variable(tf.constant(0.1, dtype=tf.float32, shape=[config.n_fc2])), 'fc3': tf.Variable(tf.constant(0.0, dtype=tf.float32, shape=[config.n_classes])) } self.cnn() def cnn(self): #x4d = tf.reshape(x, [-1, 224,224, 3]) # 卷积层 1 conv1 = tf.nn.conv2d(self.x, self.w['conv1'], strides=[1, 1, 1, 1], padding='SAME',name="conv1") conv1 = tf.nn.bias_add(conv1, self.b['conv1']) conv1 = tf.nn.relu(conv1) # 池化层 1 pool1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME') # LRN层，Local Response Normalization norm1 = tf.nn.lrn(pool1, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75) # 卷积层 2 conv2 = tf.nn.conv2d(norm1, self.w['conv2'], strides=[1, 1, 1, 1], padding='SAME') conv2 = tf.nn.bias_add(conv2, self.b['conv2']) conv2 = tf.nn.relu(conv2) # LRN层，Local Response Normalization norm2 = tf.nn.lrn(conv2, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75) # 池化层 2 pool2 = tf.nn.max_pool(norm2, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME') reshape = tf.reshape(pool2, [-1, 28* 28* 64]) dim = reshape.get_shape()[1].value fc1 = tf.add(tf.matmul(reshape, self.w['fc1']), self.b['fc1']) fc1 = tf.nn.relu(fc1) # 全连接层 2 fc2 = tf.add(tf.matmul(fc1, self.w['fc2']), self.b['fc2']) fc2 = tf.nn.relu(fc2) # 全连接层 3, 即分类层 fc2=tf.nn.dropout(fc2,self.keep_prob) self.fc3 = tf.add(tf.matmul(fc2, self.w['fc3']), self.b['fc3']) self.softmax=tf.nn.softmax(self.fc3,name="prob") # 定义损失 self.cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.fc3, labels=self.y)) self.optimizer = tf.train.AdamOptimizer(learning_rate=self.config.learning_rate).minimize(self.cost) # 评估模型 self.correct_pred = tf.equal(tf.argmax(self.fc3, 1), tf.argmax(self.y, 1)) self.accuracy = tf.reduce_mean(tf.cast(self.correct_pred, tf.float32))以上是我的卷积模型，输入图像大小为（112×112×3），每批次输入70张图片，总共迭代40次。模型结构为：两层卷积，两层池化，利用局部反应归一化和dropout来减少过拟合。最后一层利用sorftmax进行分类，损失函数为交叉熵。完整代码可以看下 github