图像分类预处理

对于训练集图片分类记录在csv文件中的情况：

有两种处理方式：

一、将训练集图片分为x_train和y_train：

#读入数据
labels=pd.read_csv('labels.csv')
sample_submission = pd.read_csv( 'sample_submission.csv')
#进行one-hot编码，并转为np.array
targets_series=pd.Series(labels['breed'])
one_hot=pd.get_dummies(targets_series,sparse=True)
one_hot_labels = np.asarray(one_hot)
#用opencv读取图片
img_size=299
x_train=[]
y_train=[]
x_test=[]
i=0
for f,breed in tqdm(labels.values):
    img=cv2.imread('train/{}.jpg'.format(f))
    label=one_hot_labels[i]
    x_train.append(cv2.resize(img,(img_size,img_size)))
    y_train.append(label)
    i+=1
for f in tqdm(sample_submission['id'].values):
    img=cv2.imread('test/{}.jpg'.format(f))
    x_test.append(cv2.resize(img,(img_size,img_size)))
#将图片处理为已有模型所需的输入
y_train_raw = np.array(y_train, np.uint8)
x_train_raw = np.array(x_train, np.float32) / 255.
x_test  = np.array(x_test, np.float32) / 255.

此方式可以直接用fit方法训练

history = model.fit(x_train_raw, 
                    y_train_raw, 
                    batch_size=16,
                    epochs=20,
                    shuffle=True,
                    validation_split=0.1)

二、将训练集图像分成训练集和开发集，按类别保存到不同的文件夹

#读取label文件
labels_csv = pd.read_csv("labels.csv")
#获取所有类别和文件名
breeds = pd.Series(labels_csv['breed'])
filenames = pd.Series(labels_csv['id'])
#将文件名与类别对应并进行one-hot编码
unique_breeds = np.unique(breeds)
labels = []
for breed in breeds:
    i = np.where(unique_breeds == breed)[0][0]
    labels.append(i)

n_breeds = np.max(labels) + 1
labels = np.eye(n_breeds)[labels]

np.where的用法

#将对应的图片拷贝到对应类的文件夹
import shutil
filenames_train = []
filenames_validate = []

# move to validate folder
for i in tqdm(range(len(filenames))):
    label = unique_breeds[np.where(labels[i]==1.)][0]
    filename = '{}.jpg'.format(filenames[i])

    if i < 8000:
        new_dir = './sorted/train/{}/'.format(label)
        filenames_train.append(new_dir + filename)
    else:
        new_dir = './sorted/validate/{}/'.format(label)
        filenames_validate.append(new_dir + filename)
        
    if not exists(new_dir):
        makedirs(new_dir)
    
    shutil.copy("train/{}.jpg".format(filenames[i]), new_dir + filename)

shuitil的用法

此方法直接用可以用ImageDataGenerator来产生训练集和开发集来训练和测试

train_dir="sorted/train/"
test_dir="sorted/validate/"
height=299
width=299
channels=3
batch_size=16
seed=1337


# Training generator
train_datagen = ImageDataGenerator(
    rotation_range = 30,
    rescale=1. / 255,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True)
train_generator = train_datagen.flow_from_directory(train_dir, 
                                                    target_size=(height,width),
                                                    batch_size=batch_size,
                                                    seed=seed,
                                                    class_mode='categorical')

# Test generator
test_datagen = ImageDataGenerator(
                                  rescale=1./255)
test_generator = test_datagen.flow_from_directory(test_dir, 
                                                  target_size=(height,width), 
                                                  batch_size=batch_size,
                                                  seed=seed,
                                                  class_mode='categorical')