def __getitem__(self, idx):
# 图片路径
batch_x = self.x_y[idx * self.batch_size:(idx + 1) * self.batch_size,
0]
# 图片标签
batch_y = self.x_y[idx * self.batch_size:(idx + 1) * self.batch_size,
1:]
# 这里是像素数组 （224，224，3）
batch_x = np.array(
[self.preprocess_img(img_path) for img_path in batch_x])
# smooth labels
batch_y = np.array(batch_y).astype(np.float32) * (1 - 0.05) + 0.05 / 40
# # 训练集数据增强
if self.is_train:
indexs = np.random.choice([0, 1, 2],
batch_x.shape[0],
replace=True,
p=[0.4, 0.4, 0.2])
mask_indexs = np.where(indexs == 1)
multi_indexs = np.where(indexs == 2)
if len(multi_indexs):
# 数据增强
multipy_batch_x = batch_x[multi_indexs]
multipy_batch_y = batch_y[multi_indexs]
train_datagenerator = self.train_datagen.flow(
multipy_batch_x,
multipy_batch_y,
batch_size=self.batch_size)
(multipy_batch_x, multipy_batch_y) = train_datagenerator.next()
batch_x[multi_indexs] = multipy_batch_x
batch_y[multi_indexs] = multipy_batch_y
if len(mask_indexs[0]):
# 随机遮挡
mask_batch_x = batch_x[mask_indexs]
mask_batch_y = batch_y[mask_indexs]
mask_batch_x = np.array(
[self.cutout_img(img) for img in mask_batch_x])
batch_x[mask_indexs] = mask_batch_x
batch_y[mask_indexs] = mask_batch_y
# 预处理
batch_x = np.array([preprocess_input(img) for img in batch_x])
# # plt 绘制图像时需要将其换成整型
# for index,label in enumerate(batch_y):
# print(np.argmax(label))
# plt.subplot(2,8,index+1)
# plt.imshow(batch_x[index].astype(int))
# plt.show()
# exit()
return batch_x, batch_y
def preprocess_img(self, img_path):
"""
image preprocessing
you can add your special preprocess method here
"""
img = Image.open(img_path)
#resize_scale = self.img_size[0] / max(img.size[:2])
#img = img.resize((int(img.size[0] * resize_scale), int(img.size[1] * resize_scale)))
img = img.resize((256, 256))
img = img.convert('RGB')
img = np.array(img)
img = img[16:16 + 224, 16:16 + 224]
img = preprocess_input(img)
#img = img[:, :, ::-1]
#img = self.center_img(img, self.img_size[0])
return img
def preprocess_img(img_path, img_size):
try:
img = Image.open(img_path)
img = img.resize((256, 256))
img = img.convert('RGB')
# img.show()
img = np.array(img)
imgs = []
for _ in range(10):
i = random.randint(0, 32)
j = random.randint(0, 32)
imgg = img[i:i + 224, j:j + 224]
imgg = preprocess_input(imgg)
imgs.append(imgg)
return imgs
except Exception as e:
print('发生了异常data_process：', e)
return 0
def preprocess_img(img_path, img_size):
"""
image preprocessing
you can add your special preprocess method here
"""
img = Image.open(img_path)
# resize_scale = img_size / max(img.size[:2])
# img = img.resize((int(img.size[0] * resize_scale), int(img.size[1] * resize_scale)))
img = img.resize((256, 256))
img = img.convert('RGB')
img = np.array(img)
imgs = []
for _ in range(10):
i = random.randint(0, 32)
j = random.randint(0, 32)
imgg = img[i:i + 224, j:j + 224]
imgg = preprocess_input(imgg)
imgs.append(imgg)
return imgs