深度学习-Tensorflow2.2-多分类{8}-多输出模型实例-20

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import pathlib
import os
import random
os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'
import IPython.display as display

gpu_ok = tf.test.is_gpu_available()
print("tf version:", tf.__version__)
print("use GPU", gpu_ok) # 判断是否使用gpu进行训练

data_dir = "F:/py/ziliao/数据集/multi-output-classification/dataset" # 定义路径

data_root = pathlib.Path(data_dir)
data_root

for item in data_root.iterdir():
    print(item) # 查看所有目录

all_image_paths = list(data_root.glob("*/*")) # 使用glob方法取出所有图片

image_count = len(all_image_paths)
image_count # 查看图片张数

all_image_paths = [str(path) for path in all_image_paths]
random.shuffle(all_image_paths)  # 对所有图片路径进行乱序

# 提取目录名称
label_names = sorted(item.name for item in data_root.glob("*/") if item.is_dir()) 
label_names

# 因为要预测颜色和种类 使用切割 set取出唯一的值 进行提取
color_label_names = set(name.split("_")[0] for name in label_names)
item_label_names = set(name.split("_")[1] for name in label_names) 

color_label_names,item_label_names  # 颜色，种类

# 编写一个对应的索引
color_label_to_index = dict((name,index)for index,name in enumerate(color_label_names))
item_label_to_index = dict((name,index)for index,name in enumerate(item_label_names))

color_label_to_index,item_label_to_index

# 对每一个图片label进行编码
all_image_labels = [pathlib.Path(path).parent.name for path in all_image_paths]
all_image_labels[:5],len(all_image_labels)

#  把颜色及物品转换成对应的序号
color_labels = [color_label_to_index[label.split("_")[0]]for label in all_image_labels]
item_labels = [item_label_to_index[label.split("_")[1]]for label in all_image_labels]
color_labels[:5],item_labels[:5]

# 绘制图片 及 对应的label
for n in range(3):
    image_index = random.choice(range(len(all_image_paths)))
    display.display(display.Image(all_image_paths[image_index],width=100,height=100))
    print(all_image_labels[image_index])
    print()

加载和格式化图像

img_path = all_image_paths[0]
img_path
img_raw = tf.io.read_file(img_path)
print(repr(img_raw)[:100]+"...")

img_tensor = tf.image.decode_image(img_raw)

print(img_tensor.shape)
print(img_tensor.dtype)

img_tensor = tf.cast(img_tensor,tf.float32)
img_tensor = tf.image.resize(img_tensor,[224,224])
img_final = img_tensor/255.0
print(img_final.shape)
print(img_final.numpy().min())
print(img_final.numpy().max())

def load_and_preporocess_image(path):
    image = tf.io.read_file(path) # 读取图片
    image = tf.image.decode_jpeg(image,channels=3) # 对图片进行解码
    image = tf.image.resize(image,[224,224]) # 定义图片形状
    image = tf.cast(image,tf.float32) # 改变图片的数据类型
    image = image/255.0 # 归一化
    image = 2*image-1  #  归一化到-1到1 之间
    return image

image_path = all_image_paths[0] # 取出第一个路径
label = all_image_labels[0] # 取出第一个label

plt.imshow((load_and_preporocess_image(image_path)+1)/2) # 使用上面函数对图片进行处理
plt.grid(False)
plt.xlabel(label)
print()

path_ds = tf.data.Dataset.from_tensor_slices(all_image_paths) # 创建样本数据集

AUTOTUNE = tf.data.experimental.AUTOTUNE
image_ds = path_ds.map(load_and_preporocess_image,num_parallel_calls=AUTOTUNE)

label_ds = tf.data.Dataset.from_tensor_slices((color_labels,item_labels)) # 创建目标值数据集

for ele in label_ds.take(3):
    print(ele[0].numpy(),ele[1].numpy())

image_label_ds = tf.data.Dataset.zip((image_ds,label_ds)) # 拼接
image_label_ds

# 划分数据集
test_count = int(image_count*0.2) # 测试集取百分之20
train_count = image_count-test_count

train_data = image_label_ds.skip(test_count) # 训练集 ： skip 跳过 test数据集
test_data = image_label_ds.take(test_count) #  测试集

BATCH_SIZE = 32 # 定义批次

train_data = train_data.shuffle(buffer_size=train_count).repeat(-1) # 对训练集进行乱序，repeat(-1) 一直重复
train_data = train_data.batch(BATCH_SIZE)
train_data = train_data.prefetch(buffer_size=AUTOTUNE)
train_data

test_data = test_data.batch(BATCH_SIZE)

建立模型

mobile_net = tf.keras.applications.MobileNetV2(input_shape=(224, 224, 3),
                                               include_top=False)

inputs = tf.keras.Input(shape=(224, 224, 3))

x = mobile_net(inputs)
x = tf.keras.layers.GlobalAveragePooling2D()(x)
x1 = tf.keras.layers.Dense(1024,activation="relu")(x)
out_color = tf.keras.layers.Dense(len(color_label_names),
                                  activation="softmax")(x1)

x2 = tf.keras.layers.Dense(1024,activation="relu")(x)
out_item = tf.keras.layers.Dense(len(item_label_names),
                                 activation="softmax")(x2)

model = tf.keras.Model(inputs = inputs,
                       outputs = [out_color,out_item])

model.summary()

# 编译
model.compile(optimizer="adam",
             loss="sparse_categorical_crossentropy", 
             metrics=["acc"])

train_steps = train_count//BATCH_SIZE
test_steps = test_count//BATCH_SIZE

model.fit(train_data,
          epochs=10,
          steps_per_epoch=train_steps,
          validation_data=test_data,
          validation_steps=test_steps)