def img_to_array(img, data_format=None, dtype=None):
if data_format is None:
data_format = backend.image_data_format()
if 'dtype' in generic_utils.getargspec(image.img_to_array).args:
if dtype is None:
dtype = backend.floatx()
return image.img_to_array(img, data_format=data_format, dtype=dtype)
return image.img_to_array(img, data_format=data_format)
def getOneStatic(imPath, sampSize, newSize):
''' randomly samples an image with no transformations '''
im = image.load_img(imPath).resize((newSize, newSize))
r = int(np.random.rand() * (newSize - sampSize))
c = int(np.random.rand() * (newSize - sampSize))
im = im.crop((r, c, r + sampSize, c + sampSize))
return image.img_to_array(im) / 255.
def data_augumentation(imgs, img_type):
from image import array_to_img, img_to_array
import cv2
tmp = list()
for x in xrange(imgs.shape[0]):
img = imgs[x]
if img_type == 'clabel':
img = clabel_to_img(img, 128, 128)
tmp.append(array_to_img(img))
augmentated_list = rotate_img(tmp)
augmentated_list = flip_img(augmentated_list)
total = len(augmentated_list)
augmentated_imgs = np.zeros([total, 1, 128, 128], dtype='float32')
for x in xrange(total):
if img_type == 'clabel':
data_path = os.path.join('test/retrain/label/img' + str(x) +
'.png')
augmentated_list[x].save(data_path)
else:
data_path = os.path.join('test/retrain/raw/img' + str(x) + '.png')
augmentated_list[x].save(data_path)
augmentated_imgs[x] = img_to_array(augmentated_list[x]) / 255
if img_type == 'raw':
return augmentated_imgs
if img_type == 'clabel':
augmentated_clabels = categorize_label(augmentated_imgs)
return augmentated_clabels
def get_divided_img_array(tmp_list):
from image import img_to_array, divide_img
imgs = [
img_to_array(patch) / 255 for tmp in tmp_list
for patch in divide_img(tmp)
]
return np.stack(imgs, axis=0)
def getImage(sampSize, newSize, imPath, _=''):
''' randomly samples an image '''
f1 = int(np.sign(np.random.rand() - .5))
f2 = int(np.sign(np.random.rand() - .5))
im = image.load_img(imPath).resize((newSize, newSize))
r = int(np.random.rand() * (newSize - sampSize))
c = int(np.random.rand() * (newSize - sampSize))
im = im.crop((r, c, r + sampSize, c + sampSize))
im = image.random_rotation(im, 5)
im = image.img_to_array(im)[:, ::-f1, ::-f2]
im = (im - im.min()) / (im.max() - im.min())
return im
def getOne(batchSize, imPath):
allIms = np.zeros((batchSize, 3, sampSize, sampSize))
for i in xrange(batchSize):
f1 = int(np.sign(np.random.rand() - .5))
f2 = int(np.sign(np.random.rand() - .5))
im = image.load_img(imPath).resize((newSize, newSize))
r = int(np.random.rand() * (newSize - sampSize))
c = int(np.random.rand() * (newSize - sampSize))
im = im.crop((r, c, r + sampSize, c + sampSize))
im = image.random_rotation(im, 5)
allIms[i, :, :, :] = image.img_to_array(im)
allIms[i, :, :, :] = allIms[i, :, ::-f1, ::-f2] / 255.
return allIms
def Gamma_correction(path, image_save_dir, aug_number):
img = load_img(path, grayscale=True) # 这是一个PIL图像
x = img_to_array(img) # 把PIL图像转换成一个numpy数组
x = x.reshape(x.shape[0], -1) # 转成二维数组
for i in range(aug_number):
gamma = np.random.uniform(low=0.7, high=1.3)
adjusted_img = (x / 255)**gamma * 255
image_name = os.path.basename(path)
prefix = image_name.split(".")[0]
save_path = os.path.join(image_save_dir,
prefix + '_aug_gamma_' + str(i) + '.jpg')
scipy.misc.imsave(save_path, adjusted_img)
def getBatch(batchSize, path): allPaths = image.list_pictures(path) allIms = np.zeros((batchSize, 3, sampSize, sampSize)) for i in xrange(batchSize): f1 = int(np.sign(np.random.rand() - .5)) f2 = int(np.sign(np.random.rand() - .5)) im = image.load_img(allPaths[int(np.random.rand() * len(allPaths))]).resize((newSize, newSize)) r = int(np.random.rand() * (newSize - sampSize)) c = int(np.random.rand() * (newSize - sampSize)) im = im.crop((r, c, r + sampSize, c + sampSize)) im = image.random_rotation(im, 5) allIms[i, :, :, :] = image.img_to_array(im) allIms[i, :, :, :] = allIms[i, :, ::-f1, ::-f2] return allIms/255.
def process_image(self, img_path, show=False):
img = image.load_img(img_path, target_size=(150, 150))
img_tensor = image.img_to_array(img) # (height, width, channels)
img_tensor = np.expand_dims(
img_tensor, axis=0
) # (1, height, width, channels), add a dimension because the model expects this shape: (batch_size, height, width, channels)
img_tensor /= 255. # imshow expects values in the range [0, 1]
if show:
plt.imshow(img_tensor[0])
plt.axis('off')
plt.show()
return self.model.predict(img_tensor)
def Noise_Injection(path, image_save_dir, aug_number):
img = load_img(path, grayscale=True) # 这是一个PIL图像
x = img_to_array(img) # 把PIL图像转换成一个numpy数组
x = x.reshape(x.shape[0], -1) # 转成二维数组
x_shape = x.shape
for i in range(aug_number):
noise = np.random.normal(loc=0.0, scale=0.01, size=x_shape)
# noise=np.random.random(size=x_shape)
x_add_noise = np.add(x, noise)
# 保存图片
image_dir = os.path.dirname(path)
image_name = os.path.basename(path)
prefix = image_name.split(".")[0]
save_path = os.path.join(image_save_dir,
prefix + '_aug_noise_' + str(i) + '.jpg')
scipy.misc.imsave(save_path, x_add_noise)
def Scaling(path, image_save_dir, aug_number):
datagen = ImageDataGenerator(zoom_range=0.3, fill_mode='constant') # 0.3
img = load_img(path, grayscale=True) # 这是一个PIL图像
x = img_to_array(img) # 把PIL图像转换成一个numpy数组
x = x.reshape((1, ) + x.shape) # 这是一个numpy数组
image_dir = os.path.dirname(path)
image_name = os.path.basename(path)
prefix = image_name.split(".")[0]
i = 1
for batch in datagen.flow(x,
batch_size=1,
save_to_dir=image_save_dir,
save_prefix=prefix + "_aug_" + 'Scal',
save_format='jpg'):
i += 1
if i > aug_number:
break # 否则生成器会退出循环
def Random_translation(path, image_save_dir, aug_number):
datagen = ImageDataGenerator(width_shift_range=0.05,
height_shift_range=0.05) # 0.2
img = load_img(path, grayscale=True) # 这是一个PIL图像
x = img_to_array(img) # 把PIL图像转换成一个numpy数组
x = x.reshape((1, ) + x.shape) # 这是一个numpy数组
image_dir = os.path.dirname(path)
image_name = os.path.basename(path)
prefix = image_name.split(".")[0]
i = 1
for batch in datagen.flow(x,
batch_size=1,
save_to_dir=image_save_dir,
save_prefix=prefix + "_aug_" + 'trans',
save_format='jpg'):
i += 1
if i > aug_number:
break # 否则生成器会退出循环
def Random_affine_transform(path, image_save_dir, aug_number):
datagen = ImageDataGenerator(
shear_range=3) # 水平或垂直投影变换,shear_range是角度范围 #5
img = load_img(path, grayscale=True) # 这是一个PIL图像
x = img_to_array(img) # 把PIL图像转换成一个numpy数组
x = x.reshape((1, ) + x.shape) # 这是一个numpy数组
image_dir = os.path.dirname(path)
image_name = os.path.basename(path)
prefix = image_name.split(".")[0]
i = 1
for batch in datagen.flow(x,
batch_size=1,
save_to_dir=image_save_dir,
save_prefix=prefix + "_aug_" + 'aff',
save_format='jpg'):
i += 1
if i > aug_number:
break # 否则生成器会退出循环
def visualize(dir_path, imgs_test_name, imgs, visualize=True):
from image import combine_img, array_to_img, img_to_array
tmp_list = list()
for img_array in imgs:
img_array *= 255
img = array_to_img(img_array, scale=False)
tmp_list.append(img)
total = imgs_test_name.shape[0]
combined_imgs = np.zeros([total, 1, imgs.shape[2]*4, imgs.shape[3]*4], dtype = 'uint8')
for x in xrange(total):
combined_img = combine_img(tmp_list[16*x:16*(x+1)])
if visualize == True:
combined_img.save(os.path.join(dir_path, imgs_test_name[x]))
print 'save label image:', x+1, '/', total
combined_imgs[x] = img_to_array(combined_img) / 255
return combined_imgs
def getBatch(batchSize, allPaths, numClass):
allIms = np.zeros((batchSize, 3, sampSize, sampSize))
Y = np.zeros((batchSize, 1)).astype('int')
rIms = np.zeros((batchSize, 1)).astype('str')
for i in xrange(batchSize):
f1 = int(np.sign(np.random.rand() - .5))
f2 = int(np.sign(np.random.rand() - .5))
rIm = allPaths[int(np.random.rand() * len(allPaths))]
rIms[i] = rIm
im = image.load_img(rIm).resize((newSize, newSize))
Y[i] = labels[rIm.split('/')[-1]]
r = int(np.random.rand() * (newSize - sampSize))
c = int(np.random.rand() * (newSize - sampSize))
im = im.crop((r, c, r + sampSize, c + sampSize))
im = image.random_rotation(im, 5)
allIms[i, :, :, :] = image.img_to_array(im)
allIms[i, :, :, :] = allIms[i, :, ::-f1, ::-f2]
if numClass > 2:
Y2 = np.zeros((batchSize, numClass)).astype('int')
for i in xrange(batchSize):
Y2[i, Y[i][0]] = 1
Y = Y2
return allIms, Y, rIms
def post(self):
image_data = self.request.files['image'][0]
np_arr = np.fromstring(image_data['body'], np.uint8)
image_np = cv2.imdecode(np_arr, cv2.IMREAD_COLOR)
image_path = '/tmp/%s.png' % str(uuid.uuid4())
cv2.imwrite(image_path, image_np)
image_f = image.load_img(image_path, target_size=(64, 64))
image_f = image.img_to_array(image_f)
image_f = np.expand_dims(image_f, axis=0)
left_or_right = self.application.model['left_or_right']
straight_layer = self.application.model['straight']
left_or_right_result = left_or_right.predict(image_f)
straight_layer_result = straight_layer.predict(image_f)
os.remove(image_path)
self.write(
json.dumps({
'left_or_right': left_or_right_result,
'straight': straight_layer_result
}))
def read_img_test(filepath, size):
img = T.load_img(os.path.join(test_dir, filepath), target_size=size)
img = T.img_to_array(img)
return img
