影像前處理可以使用CV2套件,
這裡也把keras做影像資料增強的說明也放進來
import zipfile
with zipfile.ZipFile(open('aia-cnn-classification.zip', 'rb')) as f: f.extractall(".")
讀取實體位置檔案import os
import cv2
import numpy as np
import pandas as pd
from glob import glob
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split
import tensorflow as tf
from time import time
import warnings
warnings.filterwarnings('ignore')
"""
資料夾及檔案路徑變數
"""
"""放置全部資料集的資料夾"""
data_dir_path = '//data/aia-data/may_the_4_be_with_u/where_am_i'
target_label_file_name = 'mapping.txt'
target_label_file_path = '/'.join((data_dir_path, target_label_file_name))
train_dir = '/'.join((data_dir_path, 'train'))
testset_dir = '/'.join((data_dir_path, 'testset'))
num_classes = 15
epochs = 100
batch_size = 32
img_size = 224
def load_data(Gray2RGB=False, mean_proc=False, test_size=0.25, img_size=img_size):
""" Load target labels """
with open(target_label_file_path) as f:
all_lines = [line.split(', ') for line in f.read().splitlines()]
target_labels = dict()
for line in all_lines:
target_class, target_label = line
target_labels[target_class] = target_label
""" Create training data list """
train_list = []
img_paths = []
img_labels = []
for key in target_labels.keys():
for img_path in glob('{}/{}/*.jpg'.format(train_dir, key)):
train_list.append([img_path, target_labels[key]])
img_paths.append(img_path)
img_labels.append(target_labels[key])
""" Split the list into training set and validation set """
train_img_paths, valid_img_paths, y_train, y_valid = train_test_split(img_paths, img_labels, test_size=test_size)
X_train = []
for path in train_img_paths:
img = cv2.imread(path, 0) # 0=灰階
img = cv2.resize(img, (img_size, img_size)) #壓縮圖片
if Gray2RGB == True:
img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) #imagenet是RGB要相同
img = img.astype(np.float32)
if mean_proc == 'VGG16_ImageNet':
img = img - np.array([123.68, 116.779, 103.939])
img = img[:,:,::-1] # RGB to BGR
img = (img - np.min(img)) / np.max(img)
if mean_proc == 'DenseNet':
img /= 255.
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
img = (img - mean) / std
else:
img /= 255.
img -= 0.5
img *= 2.
X_train.append(img)
X_train = np.array(X_train, dtype=np.float32)
X_valid = []
if float(test_size) != 0.:
for path in valid_img_paths:
img = cv2.imread(path, 0)
img = cv2.resize(img, (img_size, img_size))
if Gray2RGB == True:
img = cv2.cvtColor(img,cv2.COLOR_GRAY2RGB)
img = img.astype(np.float32)
if mean_proc == 'VGG16_ImageNet':
img = img - np.array([123.68, 116.779, 103.939])
img = img[:,:,::-1] # RGB to BGR
img = (img - np.min(img)) / np.max(img)
if mean_proc == 'DenseNet':
img /= 255.
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
img = (img - mean) / std
else:
img /= 255.
img -= 0.5
img *= 2.
X_valid.append(img)
X_valid = np.array(X_valid, dtype=np.float32)
if Gray2RGB == False:
X_train = np.reshape(X_train, X_train.shape+(1,))
X_valid = np.reshape(X_valid, X_valid.shape+(1,))
y_train = tf.keras.utils.to_categorical(y_train, num_classes)
y_valid = tf.keras.utils.to_categorical(y_valid, num_classes)
return X_train, y_train, X_valid, y_valid
testset_list = []
test_id_list = []
for img_path in glob('{}/*.jpg'.format(testset_dir)):
testset_list.append(img_path)
id = img_path.split('/')[-1].split('.')[0]
test_id_list.append(id)
testset_df = pd.DataFrame({'id': test_id_list, 'path': testset_list}).sort_values(by='id')
def load_test_data(Gray2RGB=False, mean_proc=False, img_size=img_size):
img_path_list = []
for img_path in glob('{}/*.jpg'.format(testset_dir)):
img_path_list.append(img_path)
X_test = []
X_id = []
for path in img_path_list:
img = cv2.imread(path, 0)
img = cv2.resize(img, (img_size, img_size))
if Gray2RGB == True:
img = cv2.cvtColor(img,cv2.COLOR_GRAY2RGB)
img = img.astype(np.float32)
if mean_proc == 'VGG16_ImageNet':
img = img - np.array([123.68, 116.779, 103.939])
img = img[:,:,::-1] # RGB to BGR
img = (img - np.min(img)) / np.max(img)
if mean_proc == 'DenseNet':
img /= 255.
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
img = (img - mean) / std
else:
img /= 255.
img -= 0.5
img *= 2.
img_id = path.split('/')[-1].split('.')[0]
X_test.append(img)
X_id.append(img_id)
X_test = np.array(X_test, dtype=np.float32)
if Gray2RGB == False:
X_test = np.reshape(X_test, X_test.shape+(1,))
return X_test, X_id
將資料清單隨機打亂
rand_seed = int(time()) data_list = shuffle(data_list, random_state=rand_seed) data_list.head(10)one hot encoding
# np.eye(10) 生成10x10對角矩陣 y_train = np.eye(10)[y_train[:]] y_test = np.eye(10)[y_test[:]]ImageDataGenerator產生器從目錄中讀取影像 使用kersa資料增強
train_datagen = ImageDataGenerator(
rescale=1./255,#像素(0~255)壓縮到[0,1]之間
rotation_range=40,#隨機旋轉角度值
width_shift_range=0.2,#水平平移(總寬度的百分比)
height_shift_range=0.2,#垂直平移
shear_range=0.2,#隨機傾斜(順時鐘傾斜角度)
zoom_range=0.2,#隨機縮放(縮小放大比率)
horizontal_flip=True,#隨機水平翻轉(非對稱才有效)
fill_mode=nearest) #旋轉平移後出現空白填補像素的方法
test_datagen = ImageDataGenerator(rescale=1./255) # 請注意！驗證資料不應該擴充!!!
train_generator = train_datagen.flow_from_directory(
train_dir, # 目標目錄
target_size=(150, 150), # 所有圖像大小調整成 150×150
batch_size=32,
class_mode='binary') # 因為使用二元交叉熵 binary_crossentropy 作為損失，所以需要二元標籤
validation_generator = test_datagen.flow_from_directory(
validation_dir,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
# 訓練
history = model.fit_generator(
train_generator,
steps_per_epoch=100,
epochs=100,
validation_data=validation_generator,
validation_steps=50)
顯示隨機擴充的訓練影像
from keras.preprocessing import image
fnames = [os.path.join(train_cats_dir, fname) for
fname in os.listdir(train_cats_dir)]
img_path = fnames[3]
img = image.load_img(img_path, target_size=(150, 150))
x = image.img_to_array(img)
x = x.reshape((1, ) + x.shape)
i = 0
for batch in datagen.flow(x, batch_size=1):
plt.figure(i)
imgplot = plt.imshow(image.array_to_img(batch[0]))
i += 1
if i % 4 == 0:
break
plt.show()
>>> keras資料增強參數設定參考 <<<
>>> CV2影像處理範例參考 <<<
Ref:
- 台灣人工智慧學校
- Deep learning 深度學習必讀：Keras 大神帶你用 Python 實作
- keras資料增強方法可以參考
沒有留言:
張貼留言