智能计算系统实验4-1-基于 VGG19 实现图像分类

#coding=utf-8 import numpy as np import struct import os import scipy.io import scipy.misc import time import tensorflow as tf from tensorflow.python.framework import graph_util os.putenv('MLU_VISIBLE_DEVICES','') IMAGE_PATH = 'data/cat1.jpg' VGG_PATH = 'imagenet-vgg-verydeep-19.mat' def net(data_path, input_image): layers = ( 'conv1_1', 'relu1_1', 'conv1_2', 'relu1_2', 'pool1', 'conv2_1', 'relu2_1', 'conv2_2', 'relu2_2', 'pool2', 'conv3_1', 'relu3_1', 'conv3_2', 'relu3_2', 'conv3_3', 'relu3_3', 'conv3_4', 'relu3_4', 'pool3', 'conv4_1', 'relu4_1', 'conv4_2', 'relu4_2', 'conv4_3', 'relu4_3', 'conv4_4', 'relu4_4', 'pool4', 'conv5_1', 'relu5_1', 'conv5_2', 'relu5_2', 'conv5_3', 'relu5_3', 'conv5_4', 'relu5_4', 'pool5', 'fc6', 'relu6', 'fc7', 'relu7', 'fc8', 'softmax' ) data = scipy.io.loadmat(data_path) weights = data['layers'][0] net = {} current = input_image for i, name in enumerate(layers): if name[:4] == 'conv': # TODO: 从模型中读取权重、偏置加数，计算卷积结果 current kernels, bias = weights[i][0][0][0][0] #print(kernels.shape,bias.shape,current.shape) # matconvnet: weights are [height, width, in_channels, out_channels] # tensorflow: weights are [in_channels, height, width, out_channels] kernels = np.transpose(kernels, (1, 0, 2, 3)) bias = bias.reshape(-1) current = _conv_layer(current, kernels, bias) elif name[:4] == 'relu': # TODO: 执行 ReLU 计算，计算结果存入 current current = tf.nn.relu(current) elif name[:4] == 'pool': # TODO: 执行 pool 计算，计算结果存入 current current = _pool_layer(current) elif name == 'softmax': # TODO: 执行 softmax 计算，计算结果存入 current current = tf.nn.softmax(current) elif name == 'fc6': # TODO: 执行全连接层计算，计算结果存入 current kernels, bias = weights[i][0][0][0][0] # matconvnet: weights are [height, width, in_channels, out_channels] # tensorflow: weights are [in_channels, height, width, out_channels] current = tf.reshape(current,[-1,current.shape[1]*current.shape[2]*current.shape[3]]) kernels = tf.reshape(kernels,[kernels.shape[0]*kernels.shape[1]*kernels.shape[2],kernels.shape[3]]) current = tf.matmul(tf.cast(current,tf.float32),kernels) + bias[0] elif name == 'fc7': kernels, bias = weights[i][0][0][0][0] kernels = tf.reshape(kernels, [kernels.shape[0] * kernels.shape[1] * kernels.shape[2], kernels.shape[3]]) current = tf.matmul(current,kernels) + bias[0] elif name == 'fc8': kernels, bias = weights[i][0][0][0][0] kernels = tf.reshape(kernels, [kernels.shape[0] * kernels.shape[1] * kernels.shape[2], kernels.shape[3]]) current = tf.matmul(current, kernels) + bias[0] net[name] = current assert len(net) == len(layers) return net def _conv_layer(input, weights, bias): # TODO: 定义卷积层的操作步骤，input 为输入张量，weights 为权重参数，bias 为偏置参数，返回计算的结果 out = tf.nn.conv2d(input=input,filter=weights,strides=[1,1,1,1],padding='SAME') out = tf.nn.bias_add(out,bias=bias) return out def _pool_layer(input): # TODO: 定义最大池化的操作步骤，input 为输入张量，返回池化操作后的计算结果 out = tf.nn.max_pool(value=input,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME') return out def preprocess(image,mean): return image - mean def load_image(path): # TODO: 使用 scipy.misc 模块读入输入图像，调用 preprocess 函数对图像进行预处理，并返回形状为（1,244,244,3）的数组 image # TODO: 修改1处 """mean = np.array([123.68, 116.779, 103.939]) image = scipy.misc.imread(path) image = scipy.misc.imresize(image,[224,224,3]) image = np.array(image).astype(np.float32) image = preprocess(image, mean) image = np.reshape(image,[1]+list(image.shape)) print(image.shape)""" image = scipy.misc.imread(path) image = scipy.misc.imresize(image, (224, 224, 3)) mean = np.array([123.68, 116.779, 103.939]) image = np.array([preprocess(image, mean)]).astype(np.float32) image = np.reshape(image, (1, 224, 224, 3)) return image if __name__ == '__main__': input_image = load_image(IMAGE_PATH) with tf.Session() as sess: img_placeholder = tf.placeholder(tf.float32, shape=(1,224,224,3), name='img_placeholder') sess.run(tf.global_variables_initializer()) # TODO: 调用 net 函数，生成 VGG19 网络模型并保存在 nets 中 nets = net(VGG_PATH, img_placeholder) #print('----nets----',nets) for i in range(10): start = time.time() # TODO: 计算 nets #print(tf.get_default_graph().get_operation_by_name('img_placeholder').outputs[0]) preds = sess.run(nets,feed_dict={img_placeholder:input_image}) end = time.time() delta_time = end - start print("processing time: %s" % delta_time) prob = preds['softmax'][0] #print(prob) top1 = np.argmax(prob) print('Classification result: id = %d, prob = %f' % (top1, prob[top1])) print("*** Start Saving Frozen Graph ***") # We retrieve the protobuf graph definition input_graph_def = sess.graph.as_graph_def() #print(input_graph_def) output_node_names = ["Softmax"] # We use a built-in TF helper to export variables to constant output_graph_def = graph_util.convert_variables_to_constants( sess, input_graph_def, output_node_names, ) # Finally we serialize and dump the output graph to the filesystem with tf.gfile.GFile("models/vgg19.pb", "wb") as f: f.write(output_graph_def.SerializeToString()) print("**** Save Frozen Graph Done ****")