deeplearning练手之 transfer learning极速实现之猫狗分类器 (keras实现)

最新推荐文章于 2024-06-26 10:03:48 发布

wudroid

最新推荐文章于 2024-06-26 10:03:48 发布

阅读量4.4k

点赞数 4

本文链接：https://blog.csdn.net/hiccuphiccup/article/details/78695664

版权

deep learning 专栏收录该内容

2 篇文章 0 订阅

订阅专栏

这是一个无聊的猫狗分类器！

这是一个transfer learning的简单试手！

原理什么乱七八糟的别人说的够多了！我看了这几个：

点击打开链接

分类器我用了keras里的ResNet50，源码是这个：

点击打开链接

所谓极速实现，是代码不超过100行啊，很爽啊

preprocess_image.py 这个是图片预处理，我用了

点击打开链接

这个库的数据，可是我的电脑处理不了那么大的数据，就随便选了3500张作为训练，500张用于测试。

每张图片规格调整到(224,224,3)，所以最后训练集维度是(3500,224,224,3)，测试集维度是(500,224,224,3)

import cv2                 # working with, mainly resizing, images
import numpy as np         # dealing with arrays
import os                  # dealing with directories
from random import shuffle # mixing up or currently ordered data that might lead our network astray in training.
from tqdm import tqdm      # a nice pretty percentage bar for tasks. Thanks to viewer Daniel BA1/4hler for this suggestion

TRAIN_DIR = 'D:/Documents/GitHub/deep-learning-models/dataset/train_wu'
TEST_DIR = 'D:/Documents/GitHub/deep-learning-models/dataset/test'
IMG_SIZE = 224


def label_img(img):
    word_label = img.split('.')[-3]
    # conversion to one-hot array [cat,dog]
    #                            [much cat, no dog]
    if word_label == 'cat': return [1,0]
    #                             [no cat, very doggo]
    elif word_label == 'dog': return [0,1]
    
def create_train_data():
    training_data = []
    for img in tqdm(os.listdir(TRAIN_DIR)):
        label = label_img(img)
        path = os.path.join(TRAIN_DIR,img)
        img = cv2.imread(path,cv2.IMREAD_COLOR)
        img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
        training_data.append([np.array(img),np.array(label)])
    shuffle(training_data)
    np.save('train_data.npy', training_data)
    return training_data

def process_test_data():
    testing_data = []
    for img in tqdm(os.listdir(TEST_DIR)):
        path = os.path.join(TEST_DIR,img)
        img_num = img.split('.')[0]
        img = cv2.imread(path,cv2.IMREAD_COLOR)
        img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
        testing_data.append([np.array(img), img_num])
        
    shuffle(testing_data)
    np.save('test_data.npy', testing_data)
    return testing_data

train_data = create_train_data()
# If you have already created the dataset:
#train_data = np.load('train_data.npy')
#test_data = process_test_data()
#test_data = np.load('test_data.npy')

train = train_data[:-500]
test = train_data[-500:]

X = np.array([i[0] for i in train]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
Y = np.array([i[1] for i in train])
test_x = np.array([i[0] for i in test]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
test_y = np.array([i[1] for i in test])

我用了keras的Resnet50(参数是imagenet的训练结果)作为transfer learning的base model，然后去掉最后几层，并重新加了了个2类输出结果的softmax层和全连接层和一个GlobalAveragePooling2D()层。

CAT.py

# -*- coding: utf-8 -*-
"""
Created on Fri Dec  1 16:22:52 2017

@author: Administrator
"""

from keras.applications.resnet50 import ResNet50
from keras.preprocessing import image
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D
from keras import backend as k
from matplotlib.pyplot import imshow
from keras.applications.imagenet_utils import preprocess_input
import imageio
from IPython.display import SVG
from keras.utils import plot_model
from keras.utils.vis_utils import model_to_dot
from keras.models import load_model
import numpy as np

# parameters have been trained with the 'imagenet' dataset (1000 classes)
base_model = ResNet50(weights='imagenet',include_top=False)
x = base_model.output
# add top layers which we will learn by our target dataset
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation= 'relu')(x)
predictions = Dense(2, activation='softmax')(x)
# new model!
model = Model(inputs = base_model.input, outputs=   predictions)

# lock the basemodel, so we can just learn the parameters of the top layers
for layer in base_model.layers:
    layer.trainable = False
    
model.compile(optimizer='rmsprop', loss='categorical_crossentropy',metrics=['accuracy'])

# step 1:train dataset(4000 samples)
model.fit(X,Y,epochs=3,batch_size = 64)

# step 2:test dataset
preds = model.evaluate(test_x, test_y)
print ("Loss = " + str(preds[0]))
print ("Test Accuracy = " + str(preds[1]))

#model.save('catdog_4000.h5')
model = load_model('catdog_4000.h5')

# step 3:test with my image
img_path = 'cat2.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
my_image = imageio.imread(img_path)
imshow(my_image)
result = model.predict(x)
if result[0][0]>result[0][1]:
    print("this is a cat!")
else:
    print("this is a dog!")


#model.summary()
#plot_model(model, to_file='model.png')
#SVG(model_to_dot(model).create(prog='dot', format='svg'))

我只训练了3个epoch（太慢了），训练集accuracy在96%左右。我们看看测试集的识别率：