def Make_PicXML_2(sample_filename,
save_pic_filename,
save_xml_filename,
aug_modes,
datasize,
start=0):
No = []
place_x = []
place_y = 0
#画像の読み込み
pais = []
for i in range(34):
filename = sample_filename + '/' + str(i) + '.jpg'
pais.append(cv.imread(filename))
sample_height = pais[0].shape[0]
sample_width = pais[0].shape[1]
for i in range(datasize):
#サイズの倍率（90～100）
magni = min(size_x / NUMBER / sample_width,
size_y / sample_height) * random.randint(90, 100) / 100
x = int(magni * sample_width)
y = int(magni * sample_height)
#牌の種類の決定
No = []
for num in range(NUMBER):
No.append(random.randint(0, 33))
#場所の決定
place_x = []
place_x.append(random.randint(0, int(size_x - x * NUMBER)))
for num in range(NUMBER):
place_x.append(place_x[0] + x * (num + 1))
place_y = random.randint(1, int(size_y - y))
#画像の生成
img = np.zeros((size_y, size_x, 3), dtype=np.uint8)
img = cv.rectangle(img, (0, 0), (size_x, size_y), (0, 128, 0),
cv.FILLED)
for num in range(NUMBER):
pai = cv.resize(pais[No[num]], (x, y))
#牌反転処理
if random.randint(0, 1) == 0:
pai = cv.flip(pai, 0)
img[place_y:place_y + y, place_x[num]:place_x[num] + x] = pai
#bb型に変換
boxes = []
for num in range(NUMBER):
boxes.append(
ia.BoundingBox(x1=place_x[num],
y1=place_y,
x2=place_x[num] + x,
y2=place_y + y,
label=dict[str(No[num])]))
bb = ia.BoundingBoxesOnImage(boxes, shape=(size_x, size_y, 3))
#アーグメーションの実行
for mode in range(len(aug_modes)):
img_aug, bb_aug = augment(img, bb, aug_modes[mode])
#保存
'''境界線を記入した画像を生成
for num in range(len(bb_aug.bounding_boxes)):
img_aug = cv.rectangle(img_aug, (int(bb_aug.bounding_boxes[num].x1), int(bb_aug.bounding_boxes[num].y1)),(int(bb_aug.bounding_boxes[num].x2),int(bb_aug.bounding_boxes[num].y2)), (0, 0, 0) , thickness=8)
'''
filename = save_pic_filename + '/' + str(start + i) + '_' + str(
mode) + '.jpg'
cv.imwrite(filename, img_aug)
#XMLファイルの生成、保存
To_Xml(str(start + i) + '_' + str(mode),
bb_aug,
save_dir=save_xml_filename)
if i % 500 == 0:
print("now complete No." + str(i))
continue
if base[0] == 'L' or base[0] == 'R':
# print(base)
boxes = np.loadtxt(label, ndmin=2)
H = img.shape[0]
W = img.shape[1]
bbs = []
for box in boxes:
x1, y1, x2, y2, l = convert_bbs_from_yolo(box[1]*W, box[2]*H, box[3]*W, box[4]*H, box[0])
bbs.append(ia.BoundingBox(x1=x1, y1=y1, x2=x2, y2=y2,label=l))
bbs=ia.BoundingBoxesOnImage(bbs, img.shape)
for repeat in range(10):
ia.seed(np.random.randint(0,1000))
np.random.seed(np.random.randint(0,1000))
seq = iaa.Sequential([
# iaa.Multiply((1.1, 1.1)),
# iaa.Fog(),
# iaa.PerspectiveTransform(scale=0.05),
# iaa.PiecewiseAffine(scale=0.015),
# iaa.ElasticTransformation(sigma=3, alpha = 3),
# iaa.Superpixels(p_replace = 0.1, n_segments=100),
# iaa.pillike.EnhanceSharpness(factor=5),
iaa.Affine(
translate_px={"x": int(np.random.normal(0,10)), "y": int(np.random.normal(0,10))},
scale=(np.random.normal(1,0.15),np.random.normal(1,0.15)),
bndbox = read_xml_annotation(XML_DIR, name)
shutil.copy(os.path.join(XML_DIR, name), AUG_XML_DIR)
shutil.copy(os.path.join(IMG_DIR, name[:-4] + '.jpg'), AUG_IMG_DIR)
for epoch in range(AUGLOOP):
seq_det = seq.to_deterministic() # 保持坐标和图像同步改变，而不是随机
# 读取图片
img = Image.open(os.path.join(IMG_DIR, name[:-4] + '.jpg'))
# sp = img.size
img = np.asarray(img)
# bndbox 坐标增强
for i in range(len(bndbox)):
bbs = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=bndbox[i][0],
y1=bndbox[i][1],
x2=bndbox[i][2],
y2=bndbox[i][3]),
],
shape=img.shape)
bbs_aug = seq_det.augment_bounding_boxes([bbs])[0]
boxes_img_aug_list.append(bbs_aug)
# new_bndbox_list:[[x1,y1,x2,y2],...[],[]]
n_x1 = int(
max(1, min(img.shape[1],
bbs_aug.bounding_boxes[0].x1)))
n_y1 = int(
max(1, min(img.shape[0],
bbs_aug.bounding_boxes[0].y1)))
n_x2 = int(
def process_json(job):
train_csv_file = job.tempdir + '/train.csv'
test_csv_file = job.tempdir + '/test.csv'
fout = open(train_csv_file, "w")
fout.write('filename,width,height,class,xmin,ymin,xmax,ymax\n')
fout.close()
fout = open(test_csv_file, "w")
fout.write('filename,width,height,class,xmin,ymin,xmax,ymax\n')
fout.close()
statefile = 'projects/' + job.project['id'] + '/state.json'
jsonbody = job.download_to_string(statefile)
state = json.loads(jsonbody)
if hasattr(job, 'jb'):
job.jb.meta['current_step_size'] = len(state['photos'])
job.jb.meta['current_step_processed'] = 0
job.jb.save_meta()
trn_sz = len(state['photos'])
test_pct_set = False
if hasattr(job, "test_pct") and job.test_pct > 0:
trn_sz = int((1 - job.test_pct) * len(state['photos']))
test_pct_set = True
ct = 0
for photo in state['photos']:
if 'children' in photo and len(photo['children']) > 0:
tags = []
process_children(photo['children'], tags)
fotopath = 'projects/' + photo['imgUrl']
logger.debug(" looking for " + fotopath)
bbox = []
data = job.download_to_data(fotopath)
img_array = np.array(bytearray(data), dtype=np.uint8)
im = cv2.imdecode(img_array, -1)
if hasattr(job, 'desired_size') and job.desired_size > 0:
# old_size is in (height, width) format
old_size = im.shape[:2]
ratio = float(job.desired_size) / max(old_size)
new_size = tuple([int(x * ratio) for x in old_size])
# new_size should be in (width, height) format
im = cv2.resize(im, (new_size[1], new_size[0]))
delta_w = job.desired_size - new_size[1]
delta_h = job.desired_size - new_size[0]
top, bottom = 0, delta_h
left, right = 0, delta_w
color = [0, 0, 0]
im = cv2.copyMakeBorder(im,
top,
bottom,
left,
right,
cv2.BORDER_CONSTANT,
value=color)
# print(im.shape)
for ch in tags:
xmin = float(ch['xmin'] * im.shape[1])
ymin = float(ch['ymin'] * im.shape[0])
xmax = float(ch['xmax'] * im.shape[1])
ymax = float(ch['ymax'] * im.shape[0])
name = ch['tag']
if name not in job.labelmap_dict:
job.labelmap_dict[name] = len(job.labelmap_dict) + 1
bbox.append(
ia.BoundingBox(x1=xmin,
y1=ymin,
x2=xmax,
y2=ymax,
label=name))
bbs = ia.BoundingBoxesOnImage(bbox, shape=im.shape)
if ct <= trn_sz:
fout = open(train_csv_file, "a")
else:
fout = open(test_csv_file, "a")
tfile = fotopath.split('.')[0] + '.tcapture'
timages = []
if job.exists(tfile):
tcapture = json.loads(job.download_to_string(tfile))
timages = tcapture['images']
for i in range(len(bbox)):
after = bbox[i]
fout.write(fotopath + ',' + str(im.shape[0]) + ',' +
str(im.shape[1]) + ',' + after.label + ',' +
str(after.x1) + ',' + str(after.y1) + ',' +
str(after.x2) + ',' + str(after.y2) + '\n')
for tim in timages:
fout.write(tim['path'] + ',' + str(im.shape[0]) + ',' +
str(im.shape[1]) + ',' + after.label + ',' +
str(after.x1) + ',' + str(after.y1) + ',' +
str(after.x2) + ',' + str(after.y2) + '\n')
if hasattr(job, 'aug') and job.aug:
with concurrent.futures.ThreadPoolExecutor(
max_workers=multiprocessing.cpu_count()) as executor:
future_img = {
executor.submit(create_image, im, job, bbs, fout): i
for i in range(job.train_samples)
}
for future in concurrent.futures.as_completed(future_img):
img = future_img[future]
try:
data = future.result()
except Exception as exc:
print(
'train augmentation generated an exception: %s'
% (exc))
if not test_pct_set:
fout.close()
fout = open(test_csv_file, "a")
with concurrent.futures.ThreadPoolExecutor(
max_workers=multiprocessing.cpu_count(
)) as executor:
future_img = {
executor.submit(create_image, im, job, bbs, fout):
i
for i in range(job.test_samples)
}
for future in concurrent.futures.as_completed(
future_img):
img = future_img[future]
try:
data = future.result()
except Exception as exc:
print(
'test augmentation generated an exception: %s'
% (exc))
fout.close()
ct += 1
if hasattr(job, 'jb'):
job.jb.meta['current_step_processed'] += 1
job.jb.save_meta()
print(job.labelmap_dict)
job.upload_file(
train_csv_file,
'corpus/' + job.name + "/" + os.path.basename(train_csv_file))
job.upload_file(
test_csv_file,
'corpus/' + job.name + "/" + os.path.basename(test_csv_file))
job.train_csv_file = 'corpus/' + job.name + \
"/" + os.path.basename(train_csv_file)
job.test_csv_file = 'corpus/' + job.name + \
"/" + os.path.basename(test_csv_file)
flag = True
# print("Error at {}, {} ymin > ymax".format(i, j))
if not flag:
writer.addObject(_class, round(xmin), round(ymin), round(xmax),
round(ymax))
aug_bounding_boxes.append(
ia.BoundingBox(x1=xmax, y1=ymax, x2=xmin, y2=ymin))
if not f:
writer.save("darkflow/Data/Annotations/auto_{}.xml".format(i))
cv2.imwrite("darkflow/Data/Images/auto_{}.jpg".format(i), image)
if aug_bounding_boxes.__len__() > 0:
for id in range(0, 50):
seq_det = seq.to_deterministic()
image1 = seq_det.augment_image(image=image)
bounding_boxes = seq_det.augment_bounding_boxes([
ia.BoundingBoxesOnImage(aug_bounding_boxes, shape=(h, w))
])[0]
writer = Writer(
"darkflow/Data/Images/auto_{}_aug_{}.jpg".format(i, id),
image1.shape[0], image1.shape[1])
for box in bounding_boxes.bounding_boxes:
writer.addObject(_class, round(box.x2), round(box.y2),
round(box.x1), round(box.y1))
if not f:
writer.save(
"darkflow/Data/Annotations/auto_{}_aug_{}.xml".format(
i, id))
cv2.imwrite(
"darkflow/Data/Images/auto_{}_aug_{}.jpg".format(
i, id), image1)
except Exception as e:
def __getitem__(self, index):
seqDet = self.augments.to_deterministic()
# --------------------------------------------------------------------------------
img_path = self.img_files[index % len(self.img_files)].rstrip()
img = cv2.imread(img_path)
# Handles images with less than three channels
while len(img.shape) != 3:
index += 1
img_path = self.img_files[index % len(self.img_files)].rstrip()
img = cv2.imread(img_path)
h, w, _ = img.shape
# --------------------------------------------------------------------------------
label_path = self.label_files[index % len(self.img_files)].rstrip()
if os.path.exists(label_path):
labels = np.loadtxt(label_path).reshape(-1, 5)
x1 = w * (labels[:, 1] - labels[:, 3] / 2)
y1 = h * (labels[:, 2] - labels[:, 4] / 2)
x2 = w * (labels[:, 1] + labels[:, 3] / 2)
y2 = h * (labels[:, 2] + labels[:, 4] / 2)
labels[:, 1] = x1
labels[:, 2] = y1
labels[:, 3] = x2
labels[:, 4] = y2
else:
labels = np.array([])
# --------------------------------------------------------------------------------
bbs = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=label[1],
y1=label[2],
x2=label[3],
y2=label[4],
label=label[0]) for label in labels
],
shape=img.shape)
augImg = seqDet.augment_images([img])[0]
augImg = np.ascontiguousarray(augImg)
augBBS = seqDet.augment_bounding_boxes([bbs])[0]
augBBS = augBBS.remove_out_of_image().clip_out_of_image()
# --------------------------------------------------------------------------------
dim_diff = np.abs(h - w)
# Upper (left) and lower (right) padding
pad1, pad2 = dim_diff // 2, dim_diff - dim_diff // 2
# Determine padding
pad = ((pad1, pad2), (0, 0),
(0, 0)) if h <= w else ((0, 0), (pad1, pad2), (0, 0))
paddedResize = iaa.Sequential([
iaa.PadToFixedSize(w + (2 * pad[1][0]),
h + (2 * pad[0][0]),
position="center",
pad_mode='constant',
pad_cval=128),
iaa.Resize({
"height": self.img_shape[0],
"width": self.img_shape[1]
})
])
resizedImg = paddedResize.augment_images([augImg])[0]
resizedBBS = paddedResize.augment_bounding_boxes([augBBS])[0]
# --------------------------------------------------------------------------------
# origDebug = bbs.draw_on_image(img, thickness=2, color=[0, 255, 0])
# augmentedDebug = augBBS.draw_on_image(augImg, thickness=2, color=[0, 0, 255])
# inputDebug = resizedBBS.draw_on_image(resizedImg, thickness=2, color=[255, 0, 0])
# print("Original: ", origDebug.shape)
# print("Augmented: ", augmentedDebug.shape)
# print("Resized: ", inputDebug.shape)
# cv2.imshow("orig", origDebug)
# cv2.imshow("augmented", augmentedDebug)
# cv2.imshow("resized", inputDebug)
# cv2.waitKey(0)
# --------------------------------------------------------------------------------
resizedH, resizedW, _ = resizedImg.shape
# Normalize
inputImg = resizedImg / 255.
# Channels-first
inputImg = np.transpose(inputImg, (2, 0, 1))
# As pytorch tensor
inputImg = torch.from_numpy(inputImg).float()
labelsArr = np.zeros((len(resizedBBS.bounding_boxes), 5),
dtype=np.float32)
for i, box in enumerate(resizedBBS.bounding_boxes):
cx = ((box.x1 + box.x2) / 2) / resizedW
cy = ((box.y1 + box.y2) / 2) / resizedH
w = (box.x2 - box.x1) / resizedW
h = (box.y2 - box.y1) / resizedH
labelsArr[i] = [box.label, cx, cy, w, h]
# Fill matrix
filledLabels = np.zeros((self.max_objects, 5))
if labels is not None:
filledLabels[range(len(labelsArr))
[:self.max_objects]] = labelsArr[:self.max_objects]
filledLabels = torch.from_numpy(filledLabels)
return img_path, inputImg, filledLabels
def bbsoi(self):
bbs = [ia.BoundingBox(x1=0, y1=1, x2=2, y2=3)]
return [ia.BoundingBoxesOnImage(bbs, shape=self.image.shape)]
def main():
# datapath为存放训练图片的地方
datapath = '/home/zhex/data/yuncong/our/'
# original_file为需要被增强的
original_file = '/home/zhex/data/yuncong/our_train.txt' # 需要被增强的训练真值txt
# aug_file只记录了新增的增强后图片的box，要得到原始+增强的所有label：cat original_file augfile>finalfile(txt拼接)
# aug_file输出是pdpd需要的格式，pytorch需要另行转换(可以拼接得到finalfile后直接将finalfile转换)
aug_file = 'augfile_our.txt'
dict_before = readlist(original_file)
new_fp = open(aug_file, 'w')
# augscene = {'Mall': 3, 'Part_B': 10, 'Part_A': 13} # 需要哪些场景，新增几倍数量的新数据
augscene = {'our': 13}
for scene in augscene:
for i in range(augscene[scene]):
for img_id in dict_before.keys():
if scene in img_id:
print(img_id)
img = Image.open(datapath + img_id)
img = np.array(img)
bbs = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=x, y1=y, x2=x + w, y2=y + h)
for [x, y, w, h] in dict_before[img_id]
],
shape=img.shape)
# 设置数据增强方式
seq = iaa.SomeOf(
(1, 3),
[
iaa.Crop(px=(0, 16)), # 裁剪
iaa.Multiply((0.7, 1.3)), # 改变色调
iaa.Affine(scale=(0.5, 0.7)), # 放射变换
iaa.GaussianBlur(sigma=(0, 1.5)), # 高斯模糊
# iaa.AddToHueAndSaturation(value=(25,-25)),
iaa.ChannelShuffle(1), # RGB三通道随机交换
iaa.ElasticTransformation(alpha=0.1),
# iaa.Grayscale(alpha=(0.2, 0.5)),
iaa.Pepper(p=0.03),
iaa.AdditiveGaussianNoise(scale=(0.03 * 255,
0.05 * 255)),
iaa.Dropout(p=(0.03, 0.05)),
iaa.Salt(p=(0.03, 0.05)),
iaa.AverageBlur(k=(1, 3)),
iaa.Add((-10, 10)),
iaa.CoarseSalt(size_percent=0.01)
])
seq_det = seq.to_deterministic(
) # 保持坐标和图像同步改变，每个batch都要调用一次，不然每次的增强都是一样的
image_aug = seq_det.augment_images([img])[0]
bbs_aug = seq_det.augment_bounding_boxes([bbs])[0]
pic_name = img_id.split('/')[-1].split('.')[0]
pic_dir = img_id.split(pic_name)[0]
if not os.path.exists(datapath + 'myaug/' + pic_dir):
os.makedirs(datapath + 'myaug/' + pic_dir)
new_img_id = 'myaug/' + pic_dir + pic_name + '_{}'.format(
i) + '.jpg'
Image.fromarray(image_aug).save(datapath + new_img_id)
new_fp = writelist(new_fp, new_img_id,
bbs_aug.bounding_boxes)
def doAugmentation(self):
configurationFile = os.path.abspath(
'.') + '/AugmenterConfigurations.xml'
# parsing configuration file
configurationTree = ET.parse(configurationFile)
configurationRoot = configurationTree.getroot()
# the directory of images to augment
imageSourceDirectory = configurationRoot.findall(
'imageSourceDirectory')[0].text
# the directory of original xml
xmlSourceDirectory = configurationRoot.findall(
'xmlSourceDirectory')[0].text
# the directory to save augmented images
imageDestinationDirectory = configurationRoot.findall(
'imageDestinationDirectory')[0].text
# mkdir if destination directory doesn't exist
if os.path.isdir(imageDestinationDirectory) is False:
os.mkdir(imageDestinationDirectory)
# load all of the augmenters described in configuration file
augmenters = configurationRoot.findall('augmenter')
# file type
# be aware of the file type, ex: jpg v.s. JPG and jpg v.s. jpeg, they are different!
fileExtension = '.jpg'
# getting addresses of images to augment
finishPartCount = 0
imageAddresses = [
imageSourceDirectory + '/' + f
for f in os.listdir(imageSourceDirectory)
if (f.__contains__(fileExtension))
]
totalImageCount = len(imageAddresses)
# counting the number of augmenters actually using
augmenterCount = 0
for a in augmenters:
if a[1].text == 'Yes':
augmenterCount += 1
# data for showing augmented samples
originalImageToShow = None
augmentedImageToShow = []
titlesToShow = []
# augmentation
augmenterHistory = [0] * len(augmenters)
self.ui.stateTextEdit.append('Starting augmentation...')
QCoreApplication.processEvents(QEventLoop.AllEvents)
for a in augmenters:
if a[1].text == 'Yes': # checking whether the user want to use this augmenter
self.ui.stateTextEdit.append('part: ' +
str(finishPartCount + 1) + '/' +
str(augmenterCount) +
' , augmenter: ' + a[0].text)
QCoreApplication.processEvents(QEventLoop.AllEvents)
# generate imgaug augmenter
seq = self.generateAugmenter(augmenters, a[0].text,
augmenterHistory)
finishImageCount = 0
for address in imageAddresses:
boundingBoxes = []
# getting target image and xml file
image = imageio.imread(address)
augmentedTree = ET.parse(
address.replace(imageSourceDirectory,
xmlSourceDirectory).replace(
fileExtension, '.xml'))
augmentedRoot = augmentedTree.getroot()
# getting bounding boxes
for obj in augmentedRoot.iter('object'):
boundingBoxes.append(
ia.BoundingBox(x1=int(obj[4][0].text),
y1=int(obj[4][1].text),
x2=int(obj[4][2].text),
y2=int(obj[4][3].text),
label=obj[0].text))
BBS = ia.BoundingBoxesOnImage(boundingBoxes,
shape=image.shape)
# augmentation
augmentedImage, augmentedBBS = seq(image=image,
bounding_boxes=BBS)
# xml handling
for fileName in augmentedRoot.iter('filename'):
newNameAttributes = '_'
if a[0].text == 'Crop':
newNameAttributes = newNameAttributes + a[0].text
if a[2][0].text == '0' and a[2][3].text == '0':
newNameAttributes = newNameAttributes + '=' + 'topLeft'
elif a[2][0].text == '0' and a[2][1].text == '0':
newNameAttributes = newNameAttributes + '=' + 'topRight'
elif a[2][2].text == '0' and a[2][1].text == '0':
newNameAttributes = newNameAttributes + '=' + 'bottomRight'
elif a[2][2].text == '0' and a[2][3].text == '0':
newNameAttributes = newNameAttributes + '=' + 'bottomLeft'
else:
newNameAttributes = newNameAttributes + '=' + 'center_percent' + a[
2][0].text.replace('.', 'p')
elif a[0].text == 'Flipud' or a[0].text == 'Fliplr':
newNameAttributes = newNameAttributes + a[0].text
elif a[0].text == 'Rot':
newNameAttributes = newNameAttributes + a[
0].text + '=' + a[2][0].text
elif a[0].text == 'Add':
newNameAttributes = newNameAttributes + a[
0].text + '=' + a[2][0].text
else:
newNameAttributes = newNameAttributes + a[0].text
for args in a[2]:
newNameAttributes = newNameAttributes + '_' + args.tag + '=' + args.text.replace(
'.', 'p')
newName = str(
fileName.text.replace(
fileExtension,
newNameAttributes)) + fileExtension
fileName.text = newName
for folder in augmentedRoot.iter('folder'):
newFolder = str(a[0].text)
folder.text = newFolder
for path in augmentedRoot.iter('path'):
newPath = str(imageDestinationDirectory
) + '/' + a[0].text + '/' + newName
path.text = newPath
for size in augmentedRoot.iter('size'):
size[0].text = str(augmentedImage.shape[1])
size[1].text = str(augmentedImage.shape[0])
size[2].text = str(augmentedImage.shape[2])
i = 0
for obj in augmentedRoot.findall('object'):
if augmentedBBS.bounding_boxes[
i].is_fully_within_image(augmentedImage.shape):
obj[4][0].text = str(
augmentedBBS.bounding_boxes[i].x1)
obj[4][1].text = str(
augmentedBBS.bounding_boxes[i].y1)
obj[4][2].text = str(
augmentedBBS.bounding_boxes[i].x2)
obj[4][3].text = str(
augmentedBBS.bounding_boxes[i].y2)
elif augmentedBBS.bounding_boxes[
i].is_partly_within_image(
augmentedImage.shape):
augmentedBBS.bounding_boxes[i].clip_out_of_image(
augmentedImage.shape)
obj[4][0].text = str(
augmentedBBS.bounding_boxes[i].x1)
obj[4][1].text = str(
augmentedBBS.bounding_boxes[i].y1)
obj[4][2].text = str(
augmentedBBS.bounding_boxes[i].x2)
obj[4][3].text = str(
augmentedBBS.bounding_boxes[i].y2)
else:
augmentedRoot.remove(obj)
i += 1
# draw bounding boxes on image (uncomment to enable)
# augmentedImage = augmentedBBS.draw_on_image(
# augmentedImage, size=5, color=[0, 0, 255])
# saving new xml file
# mkdir if destination directory doesn't exist
if os.path.isdir(newPath.replace(newName, '')) is False:
os.mkdir(newPath.replace(newName, ''))
os.mkdir(newPath.replace(newName, '/xmlFile'))
augmentedTree.write(
newPath.replace(newName,
'/xmlFile/' + newName).replace(
fileExtension, '.xml')
) # for saving jpg and xml into seperate directories
# saving new image
imageio.imsave(newPath, augmentedImage)
# collecting samples to show
if configurationRoot.findall(
'showAugmentedSamples')[0].text == 'Yes':
if finishPartCount is 0 and finishImageCount is 0:
originalImageToShow = image
if finishImageCount is 0: # sample images to show later
augmentedImageToShow.append(augmentedImage)
titlesToShow.append(newNameAttributes[1:])
finishImageCount += 1
self.ui.stateTextEdit.append(
str(finishImageCount) + '/' + str(totalImageCount) +
', saved to: ' + newPath)
QCoreApplication.processEvents(QEventLoop.AllEvents)
finishPartCount += 1
self.ui.stateTextEdit.append('')
QCoreApplication.processEvents(QEventLoop.AllEvents)
# showing augmented samples
# checking whether the user want to show augmented samples
if configurationRoot.findall('showAugmentedSamples')[0].text == 'Yes':
totalSampleCount = math.ceil(augmenterCount / 3)
for i in range(0, augmenterCount, 3):
self.ui.stateTextEdit.append('showing sample: ' +
str(int((i / 3) + 1)) + '/' +
str(totalSampleCount))
QCoreApplication.processEvents(QEventLoop.AllEvents)
plt.figure(figsize=(12.8, 7.2))
plt.axis('off')
if augmenterCount - i is 1:
plt.subplot(1, 2, 1)
elif augmenterCount - i is 2:
plt.subplot(1, 3, 1)
else:
plt.subplot(2, 2, 1)
plt.imshow(originalImageToShow)
plt.title('Original')
j = 0
while j < 3 and i + j < len(augmentedImageToShow):
if augmenterCount - i is 1:
plt.subplot(1, 2, j + 2)
elif augmenterCount - i is 2:
plt.subplot(1, 3, j + 2)
else:
plt.subplot(2, 2, j + 2)
plt.imshow(augmentedImageToShow[i + j])
plt.title(titlesToShow[i + j])
plt.axis('off')
j += 1
plt.show()
self.ui.stateTextEdit.append('Augmentation finished.')
QCoreApplication.processEvents(QEventLoop.AllEvents)
import imgaug
from skimage.data import chelsea
if __name__ == '__main__':
raw_im = chelsea()
im = imgaug.imresize_single_image(raw_im, raw_im.shape[:2])
# j,i,j+,i+
bb = imgaug.BoundingBox(128, 64, 128 + 64 + 160, 64 + 128)
imgaug.imshow(bb.draw_on_image(im, color=(0, 255, 0), size=5))
seq = imgaug.augmenters.Sequential([
imgaug.augmenters.Resize((0.5, 2.)),
imgaug.augmenters.Fliplr(),
imgaug.augmenters.Flipud(),
imgaug.augmenters.CropToFixedSize(width=128, height=128)
])
for _ in range(10):
aug_ims, aug_bbs = seq(image=im,
bounding_boxes=imgaug.BoundingBoxesOnImage(
[bb], shape=im.shape))
imgaug.imshow(aug_bbs.draw_on_image(aug_ims, color=(0, 255, 0),
size=5))
def data_aug(img, bboxs=None, keypoints=None, joint_nums=17):
'''
:param img: 需要进行数据增强的图像
:param bboxs: list, [ [x1, y1, x2, y2], ..., [xn1, yn1, xn2, yn2] ]
:param keypoints: 关键点, COCO format or Ai-challenger format, list of list, [ [num_joints x 3], [num_joints x 3], ..., ]
:return:
'''
is_flip = [random.randint(0, 1), random.randint(0, 1)]
seq = iaa.Sequential([
iaa.Affine(rotate=(-15, 15), scale=(0.8, 1.2), mode='constant'),
iaa.Multiply((0.7, 1.5)),
iaa.Grayscale(iap.Choice(a=[0, 1], p=[0.8, 0.2]),
from_colorspace='BGR'),
iaa.Fliplr(is_flip[0]),
iaa.Flipud(is_flip[1]),
])
seq_det = seq.to_deterministic()
bbs = None
kps = None
bbs_aug = None
kps_aug = None
# joint_nums = 14
new_bboxs = []
new_keypoints = []
kps_ori = copy.copy(keypoints)
kps_ori = np.reshape(np.asarray(kps_ori),
newshape=(-1, joint_nums,
3)) if kps_ori is not None else None
if bboxs is not None:
assert type(bboxs) == type([])
bbs = ia.BoundingBoxesOnImage([], shape=img.shape)
for box in bboxs:
bbs.bounding_boxes.append(
ia.BoundingBox(x1=box[0], y1=box[1], x2=box[2], y2=box[3]))
if keypoints is not None:
kps = ia.KeypointsOnImage([], shape=img.shape)
assert type(keypoints) == type([])
for single_person_keypoints in keypoints:
for i in range(joint_nums):
joint = single_person_keypoints[i * 3:i * 3 + 3]
kps.keypoints.append(ia.Keypoint(x=joint[0], y=joint[1]))
img_aug = seq_det.augment_image(img)
if bbs is not None:
# print(bbs)
bbs_aug = seq_det.augment_bounding_boxes([bbs])
# print(bbs_aug)
for i in range(len(bbs_aug[0].bounding_boxes)):
box_aug = bbs_aug[0].bounding_boxes[i]
box = [box_aug.x1, box_aug.y1, box_aug.x2, box_aug.y2]
new_bboxs.append(box)
if kps is not None:
# print(kps)
kps_aug = seq_det.augment_keypoints([kps])
for i in range(len(kps_aug[0].keypoints)):
point = kps_aug[0].keypoints[i]
new_keypoints.append([point.x, point.y, 1])
new_keypoints = np.reshape(np.asarray(new_keypoints),
newshape=(-1, joint_nums, 3))
# keep ori keypoint visiable attribute
for i in range(kps_ori.shape[0]):
for joint in range(kps_ori.shape[1]):
new_keypoints[i][joint][2] = kps_ori[i][joint][2]
if kps_ori[i][joint][0] == 0 or kps_ori[i][joint][1] == 0:
new_keypoints[i][joint] = np.asarray([0, 0, 0])
# if flip, change keypoint order (left <-> right)
# ai-format: [ 0-right_shoulder, 1-right_elbow, 2-right_wrist,
# 3-left_shoulder, 4-left_elbow, 5-left_wrist,
# 6-right_hip, 7-right_knee, 8-right_ankle,
# 9-left_hip, 10-left_knee, 11-left_ankle,
# 12-head, 13-neck ]
# coco-format: TODO add coco-foramt change index
# change_index = [[0, 3], [1, 4], [2, 5], [6, 9], [7, 10], [8, 11]]
change_index = [[1, 2], [3, 4], [5, 6], [7, 8], [9, 10], [11, 12],
[13, 14], [15, 16]]
for flip in is_flip:
if flip:
for i in range(kps_ori.shape[0]):
for index in change_index:
right_point = copy.copy(new_keypoints[i][index[0]])
new_keypoints[i][index[0]] = new_keypoints[i][index[1]]
new_keypoints[i][index[1]] = right_point
new_keypoints = [
list(np.reshape(single_person_keypoints, (-1, )))
for single_person_keypoints in new_keypoints
]
# test
# if bbs is not None:
# img_before = bbs.draw_on_image(img, color=(0, 255, 0), thickness=2)
# img_after = bbs_aug.draw_on_image(img_aug, color=(0,0,255), thickness=2)
# cv2.imshow('box ori', img_before)
# cv2.imshow('box after', img_after)
# cv2.waitKey(0)
# if kps is not None:
# img_before = kps.draw_on_image(img, color=(0, 255, 0), size=5)
# img_after = kps_aug.draw_on_image(img_aug, color=(0, 0, 255), size=5)
# for i in range(kps_ori.shape[0]):
# for joint in range(kps_ori.shape[1]):
# point = kps_ori[i][joint]
# cv2.putText(img_before, str(joint), (int(point[0]), int(point[1])), cv2.FONT_HERSHEY_COMPLEX, 1, (0, 0, 250), 1)
# point = new_keypoints[i][3*joint:3*joint+3]
# # cv2.putText(img_after, str(point[2]), (int(point[0]), int(point[1])), cv2.FONT_HERSHEY_COMPLEX, 1, (0, 0, 250), 1)
# cv2.putText(img_after, str(joint), (int(point[0]), int(point[1])), cv2.FONT_HERSHEY_COMPLEX, 1,
# (0, 0, 250), 1)
# cv2.imshow('kps ori', img_before)
# cv2.imshow('kps after', img_after)
# cv2.waitKey(0)
return img_aug, new_bboxs, new_keypoints
def main():
image = data.astronaut()
image = ia.imresize_single_image(image, (HEIGHT, WIDTH))
kps = []
for y in range(NB_ROWS):
ycoord = BB_Y1 + int(y * (BB_Y2 - BB_Y1) / (NB_COLS - 1))
for x in range(NB_COLS):
xcoord = BB_X1 + int(x * (BB_X2 - BB_X1) / (NB_ROWS - 1))
kp = (xcoord, ycoord)
kps.append(kp)
kps = set(kps)
kps = [ia.Keypoint(x=xcoord, y=ycoord) for (xcoord, ycoord) in kps]
kps = ia.KeypointsOnImage(kps, shape=image.shape)
bb = ia.BoundingBox(x1=BB_X1, x2=BB_X2, y1=BB_Y1, y2=BB_Y2)
bbs = ia.BoundingBoxesOnImage([bb], shape=image.shape)
pairs = []
params = [{
"rotate": 45
}, {
"translate_px": 20
}, {
"translate_percent": 0.1
}, {
"scale": 1.2
}, {
"scale": 0.8
}, {
"shear": 45
}, {
"rotate": 45,
"cval": 255
}, {
"translate_px": 20,
"mode": "constant"
}, {
"translate_px": 20,
"mode": "edge"
}, {
"translate_px": 20,
"mode": "symmetric"
}, {
"translate_px": 20,
"mode": "reflect"
}, {
"translate_px": 20,
"mode": "wrap"
}, {
"scale": 0.5,
"order": 0
}, {
"scale": 0.5,
"order": 1
}, {
"scale": 0.5,
"order": 2
}, {
"scale": 0.5,
"order": 3
}, {
"scale": 0.5,
"order": 4
}, {
"scale": 0.5,
"order": 5
}, {
"rotate": 45,
"translate_px": 20,
"scale": 1.2
}, {
"rotate": 45,
"translate_px": 20,
"scale": 0.8
}, {
"rotate": (-45, 45),
"translate_px": (-20, 20),
"scale": (0.8, 1.2),
"order": ia.ALL,
"mode": ia.ALL,
"cval": ia.ALL
}, {
"rotate": (-45, 45),
"translate_px": (-20, 20),
"scale": (0.8, 1.2),
"order": ia.ALL,
"mode": ia.ALL,
"cval": ia.ALL
}, {
"rotate": (-45, 45),
"translate_px": (-20, 20),
"scale": (0.8, 1.2),
"order": ia.ALL,
"mode": ia.ALL,
"cval": ia.ALL
}, {
"rotate": (-45, 45),
"translate_px": (-20, 20),
"scale": (0.8, 1.2),
"order": ia.ALL,
"mode": ia.ALL,
"cval": ia.ALL
}]
seqs_skimage = [iaa.Affine(backend="skimage", **p) for p in params]
seqs_cv2 = [iaa.Affine(backend="auto", **p) for p in params]
#seqs_cv2 = []
#for p in params:
# seqs_cv2 = [iaa.Affine(backend="cv2", **p) for p in params]
for seq_skimage, seq_cv2 in zip(seqs_skimage, seqs_cv2):
#seq_skimage.localize_random_state_()
#seq_cv2.localize_random_state_()
#seq_cv2.copy_random_state_(seq_skimage)
seq_skimage_det = seq_skimage.to_deterministic()
seq_cv2_det = seq_cv2.to_deterministic()
seq_cv2_det.copy_random_state_(seq_skimage_det)
image_aug_skimage = seq_skimage_det.augment_image(image)
image_aug_cv2 = seq_cv2_det.augment_image(image)
#print(image_aug.dtype, np.min(image_aug), np.max(image_aug))
kps_aug_skimage = seq_skimage_det.augment_keypoints([kps])[0]
kps_aug_cv2 = seq_cv2_det.augment_keypoints([kps])[0]
bbs_aug_skimage = seq_skimage_det.augment_bounding_boxes([bbs])[0]
bbs_aug_cv2 = seq_cv2_det.augment_bounding_boxes([bbs])[0]
image_before_skimage = np.copy(image)
image_before_cv2 = np.copy(image)
image_before_skimage = kps.draw_on_image(image_before_skimage)
image_before_cv2 = kps.draw_on_image(image_before_cv2)
image_before_skimage = bbs.draw_on_image(image_before_skimage)
image_before_cv2 = bbs.draw_on_image(image_before_cv2)
image_after_skimage = np.copy(image_aug_skimage)
image_after_cv2 = np.copy(image_aug_cv2)
image_after_skimage = kps_aug_skimage.draw_on_image(
image_after_skimage)
image_after_cv2 = kps_aug_cv2.draw_on_image(image_after_cv2)
image_after_skimage = bbs_aug_skimage.draw_on_image(
image_after_skimage)
image_after_cv2 = bbs_aug_cv2.draw_on_image(image_after_cv2)
pairs.append(
np.hstack(
(image_before_skimage, image_after_skimage, image_after_cv2)))
misc.imshow(np.vstack(pairs))
misc.imsave("affine.jpg", np.vstack(pairs))
img_lst = glob(current_dir + "/*.jpg")
for img_name in tqdm(img_lst):
img = Image.open(img_name)
img = np.array(img)
results = readAnnotations(img_name.replace("jpg", "xml"))
bndboxes = []
for bndbox in results:
bndboxes.append(
ia.BoundingBox(x1=bndbox[1],
y1=bndbox[2],
x2=bndbox[3],
y2=bndbox[4]))
bbs = ia.BoundingBoxesOnImage(bndboxes, shape=img.shape)
for i in range(int(args.times)):
seq = iaa.Sequential([
iaa.Flipud(0.5),
iaa.Multiply((1.2, 1.5)),
iaa.Affine(translate_px={
"x": 10,
"y": 10
},
scale=(0.8, 0.95),
rotate=(-10, 10))
])
# 保持坐标和图像同步改变，而不是随机
seq_det = seq.to_deterministic()
def __call__(self, sample):
image, keypoints, bboxes, is_visible, size ,name = sample['image'], sample['keypoints'], sample['bbox'], sample['is_visible'], sample['size'], sample['name']
h, w = image.shape[:2]
kps_coords = []
kps = []
for temp in keypoints:
if temp[0] !=0 or temp[1] !=0:
kps_coords.append((temp[0],temp[1]))
for kp_x, kp_y in kps_coords:
kps.append(ia.Keypoint(x=kp_x, y=kp_y))
box_list = []
for bbox in bboxes:
box_list.append(ia.BoundingBox(x1 = bbox[0]-bbox[2]//2,
y1=bbox[1]-bbox[3]//2,
x2=bbox[0]+bbox[2]//2,
y2=bbox[1]+bbox[3]//2))
bbs = ia.BoundingBoxesOnImage(box_list, shape=image.shape)
kps_oi = ia.KeypointsOnImage(kps, shape=image.shape)
if self.mode =='train':
seq = iaa.Sequential([
iaa.Affine(
rotate=(-40, 40),
scale=(0.35, 2.5),
), # random rotate by -40-40deg and scale to 35-250%, affects keypoints
iaa.Crop(
px = (int(0.1*random.random()*h), int(0.1*random.random()*w), int(0.1*random.random()*h), int(0.1*random.random()*w)),
),
iaa.Resize({"height": self.output_size[0], "width": self.output_size[1]})
])
else:
seq = iaa.Sequential([
iaa.Scale({"height": self.output_size[0], "width": self.output_size[1]})
])
seq_det = seq.to_deterministic()
image_aug = seq_det.augment_images([image])[0]
keypoints_aug = seq_det.augment_keypoints([kps_oi])[0]
bbs_aug = seq_det.augment_bounding_boxes([bbs])[0]
# update keypoints and bbox
cnt = 0
for ck, temp in enumerate(keypoints):
if temp[0] != 0 or temp[1] != 0:
# ignore outside keypoints
if keypoints_aug.keypoints[cnt].x >=0 and keypoints_aug.keypoints[cnt].x < image_aug.shape[1] and \
keypoints_aug.keypoints[cnt].y >=0 and keypoints_aug.keypoints[cnt].y < image_aug.shape[0]:
temp[0] = keypoints_aug.keypoints[cnt].x
temp[1] = keypoints_aug.keypoints[cnt].y
else:
temp[0] = 0.0
temp[1] = 0.0
cnt +=1
keypoints = np.asarray(keypoints, dtype= np.float32)
# Delete empty keypoints
keypoints = list(keypoints.reshape(-1,len(name_list),2))
blacklist = []
for idx, (temp, tempbb, tempvis) in enumerate(zip(keypoints, bbs_aug.bounding_boxes, is_visible)):
if tempbb.x1 < 0.0:
tempbb.x1 = 0.0
elif tempbb.x1 > image_aug.shape[1]:
tempbb.x1 = image_aug.shape[1]
if tempbb.y1 < 0.0:
tempbb.y1 = 0.0
elif tempbb.y1 > image_aug.shape[0]:
tempbb.y1 = image_aug.shape[0]
if tempbb.x2 < 0.0:
tempbb.x2 = 0.0
elif tempbb.x2 > image_aug.shape[1]:
tempbb.x2 = image_aug.shape[1]
if tempbb.y2 < 0.0:
tempbb.y2 = 0.0
elif tempbb.y2 > image_aug.shape[0]:
tempbb.y2 = image_aug.shape[0]
if (np.unique(temp) == 0).all():
blacklist.append(idx)
# Update keypoint visibility
for jdx, (yx, vis) in enumerate(zip(list(temp), tempvis)):
if (np.unique(yx) == 0).all():
tempvis[jdx] = False
for i in sorted(blacklist, reverse=True):
del keypoints[i]
del bbs_aug.bounding_boxes[i]
del is_visible[i]
if len(keypoints) == 0:
keypoints.append(np.zeros((len(name_list),2) ))
keypoints = np.asarray(keypoints, dtype= np.float32)
# bbox
new_bboxes = []
if len(bbs_aug.bounding_boxes) > 0:
for i in range(len(bbs_aug.bounding_boxes)):
new_bbox = []
temp = bbs_aug.bounding_boxes[i]
new_bbox.append((temp.x2+temp.x1)/2) #center x
new_bbox.append((temp.y2+temp.y1)/2) #center y
new_bbox.append((temp.x2-temp.x1)) #width
new_bbox.append((temp.y2-temp.y1)) #height
new_bboxes.append(new_bbox)
return {'image': image_aug, 'keypoints': keypoints, 'bbox': new_bboxes, 'is_visible':is_visible, 'size': size}
dir = 'images_augmentation/train/'
dir_original = 'images_augmentation/train_original/'
images, annotations = read_train_dataset(dir_original)
for idx in range(len(images)):
image = images[idx]
boxes = annotations[idx][0]
ia_bounding_boxes = []
for box in boxes:
ia_bounding_boxes.append(
ia.BoundingBox(x1=box[1], y1=box[2], x2=box[3], y2=box[4]))
bbs = ia.BoundingBoxesOnImage(ia_bounding_boxes, shape=image.shape)
seq = iaa.Sequential([
#iaa.Multiply((1.2, 1.5)),
iaa.Affine(translate_px={
"x": -30,
"y": -40
},
scale=(0.6, 0.8)
#rotate=-30
)
])
seq_det = seq.to_deterministic()
image_aug = seq_det.augment_images([image])[0]
transf = iaa.GaussianBlur(2.5)
elif da_choice==3:
transf=iaa.ElasticTransformation(alpha=(0.5, 1.5), sigma=0.25)
elif da_choice==4:
transf=iaa.ContrastNormalization((0.5, 2.0))
elif da_choice==5:
transf=iaa.Sharpen(alpha=(0,1.0),lightness=20.0)
elif da_choice==6:
transf=iaa.Emboss(alpha=(0, 1.0), strength=(0, 2.0))
elif da_choice==7:
transf=iaa.SaltAndPepper(0.08)
if da==1:
#Data augmentation
imcv2=np.array(im)
bb=ia.BoundingBoxesOnImage([ia.BoundingBox(x1=x_min, y1=y_min, x2=x_max, y2=y_max)],shape=imcv2.shape)
transf_det=transf.to_deterministic()
bbs_aug=transf_det.augment_bounding_boxes([bb])[0]
bbs_aug=bbs_aug.bounding_boxes[0]
x_min=bbs_aug.x1
x_max=bbs_aug.x2
y_min=bbs_aug.y1
y_max=bbs_aug.y2
print(x_min)
print(x_max)
print(y_min)
print(y_max)
xmn.text=str(np.float64((x_min)/1)) #/widthc
def image_augment(seq, data_dir, out_dir, out_label_dir, annotation_xml, name,
classes):
boxes_img_aug_list = []
imgaug_files = []
new_bndbox_list = []
bndbox = read_xml_annotation(annotation_xml, name)
for epoch in range(FLAGS.imgaugloop):
# Keep the coordinates and the image changing synchronously, not randomly
seq_det = seq.to_deterministic()
# read image
img = Image.open(
os.path.join(data_dir, "JPEGImages", name[:-4] + ".jpg"))
img = np.array(img)
# change bndbox
for i in range(len(bndbox)):
bbs = ia.BoundingBoxesOnImage(
[
ia.BoundingBox(
x1=bndbox[i][0],
y1=bndbox[i][1],
x2=bndbox[i][2],
y2=bndbox[i][3],
),
],
shape=img.shape,
)
bbs_aug = seq_det.augment_bounding_boxes([bbs])[0]
boxes_img_aug_list.append(bbs_aug)
# new_bndbox_list:[[x1,y1,x2,y2],...[],[]]
new_bndbox_list.append([
int(bbs_aug.bounding_boxes[0].x1),
int(bbs_aug.bounding_boxes[0].y1),
int(bbs_aug.bounding_boxes[0].x2),
int(bbs_aug.bounding_boxes[0].y2),
])
# save the augment image
image_aug = seq_det.augment_images([img])[0]
auged_name = str(name[:-4]) + "_aug_" + str(epoch) + ".jpg"
path = os.path.join(data_dir, "JPEGImages", auged_name)
image_auged = bbs.draw_on_image(image_aug, thickness=0)
image_saved = Image.fromarray(image_auged)
image_saved.save(path)
# draw bbox
if FLAGS.augcheck:
draw = ImageDraw.Draw(image_saved)
for rect in new_bndbox_list:
draw.rectangle(rect, outline=(0, 255, 0), width=2)
val_path = os.path.join(
data_dir,
"JPEGImages",
str(name[:-4]) + "_aug_" + str(epoch) + "_val.jpg",
)
image_saved.save(val_path)
# save new XML
auged_xml = change_xml_list_annotation(
annotation_xml,
name[:-4],
new_bndbox_list,
os.path.join(data_dir, "Annotations"),
epoch,
)
# print(auged_name)
convert_annotation(data_dir, out_dir, out_label_dir, auged_xml,
auged_name, classes)
# append to imgaug_files
imgaug_files.append(
os.path.relpath(os.path.join(out_label_dir, auged_name)))
new_bndbox_list = []
return imgaug_files
def doAugmentation():
# argparse
parser = ArgumentParser()
parser.add_argument("conf", help="the path of augmenter configurations")
arguments = parser.parse_args()
configurationFile = os.path.abspath('.') + '/' + arguments.conf
# for debugging
# configurationFile = os.path.abspath('.') + '\\AugmenterConfigurations.xml'
# parsing configuration file
configurationTree = ET.parse(configurationFile)
configurationRoot = configurationTree.getroot()
# the directory of images to augment
imageSourceDirectory = configurationRoot.findall(
'imageSourceDirectory')[0].text
# the directory of original xml
xmlSourceDirectory = configurationRoot.findall(
'xmlSourceDirectory')[0].text
# the directory to save augmented images
imageDestinationDirectory = configurationRoot.findall(
'imageDestinationDirectory')[0].text
# the directory to save augmented xml files
xmlDestinationDirectory = configurationRoot.findall(
'xmlDestinationDirectory')[0].text
# mkdir if destination directory doesn't exist
if os.path.isdir(imageDestinationDirectory) is False:
os.mkdir(imageDestinationDirectory)
'''deprecated
# augmenters. Visit https://github.com/aleju/imgaug to find more augmenters.
# sequential augmentation. Applies all the augmenters(sequentially) to an image, generates an image that contains all the augmenters.
# seq = iaa.Sequential([
# iaa.SigmoidContrast(cutoff=0.6, gain=7)]) # modify "iaa.Add(-45), iaa.Crop(precent=0.2), iaa.GaussianBlur(2)" to get the augmentation type you want
# seperate augmentation. Applies one augmenter to an image and generated augmented image.
deprecated'''
# load all of the augmenters in configuration file
augmenters = configurationRoot.findall('augmenter')
# file type
# be aware of the file type, ex: jpg vs. JPG or jpg vs. jpeg, they are different!
fileExtension = configurationRoot.findall('fileExtension')[0].text
# getting addresses of images to augment
imageAddresses = []
totalImageCount = 0
finishPartCount = 0
for f in os.listdir(imageSourceDirectory):
if (f.__contains__(fileExtension)):
imageAddresses.append(imageSourceDirectory + '/' + f)
totalImageCount += 1
# counting the number of augmenters actually using
augmenterCount = 0
for a in augmenters:
if a[1].text == 'Yes':
augmenterCount += 1
# data for showing augmented samples
originalImageToShow = None
augmentedImageToShow = []
titlesToShow = []
# augmentation
augmenterHistory = [0] * len(augmenters)
print("starting augmentation")
for a in augmenters:
if a[1].text == 'Yes': # checking whether the user want to use this augmenter
print("part: " + str(finishPartCount + 1) + "/" +
str(augmenterCount))
# generate imgaug augmenter
seq = generateAugmenter(augmenters, a[0].text, augmenterHistory)
finishImageCount = 0
for address in imageAddresses:
boundingBoxes = []
# getting target image and xml file
image = imageio.imread(address)
augmentedTree = ET.parse(
address.replace(imageSourceDirectory,
xmlSourceDirectory).replace(
fileExtension, '.xml'))
augmentedRoot = augmentedTree.getroot()
# getting bounding boxes
for obj in augmentedRoot.iter('object'):
boundingBoxes.append(
ia.BoundingBox(x1=int(obj[4][0].text),
y1=int(obj[4][1].text),
x2=int(obj[4][2].text),
y2=int(obj[4][3].text),
label=obj[0].text))
BBS = ia.BoundingBoxesOnImage(boundingBoxes, shape=image.shape)
# augmentation
augmentedImage, augmentedBBS = seq(image=image,
bounding_boxes=BBS)
# xml handling
for fileName in augmentedRoot.iter('filename'):
newNameAttributes = '_'
if a[0].text == 'Crop':
newNameAttributes = newNameAttributes + a[0].text
if a[2][0].text == '0' and a[2][3].text == '0':
newNameAttributes = newNameAttributes + '=' + 'topLeft'
elif a[2][0].text == '0' and a[2][1].text == '0':
newNameAttributes = newNameAttributes + '=' + 'topRight'
elif a[2][2].text == '0' and a[2][1].text == '0':
newNameAttributes = newNameAttributes + '=' + 'bottomRight'
elif a[2][2].text == '0' and a[2][3].text == '0':
newNameAttributes = newNameAttributes + '=' + 'bottomLeft'
elif a[0].text == 'Flipud' or a[0].text == 'Fliplr':
newNameAttributes = newNameAttributes + a[0].text
elif a[0].text == 'Rot':
newNameAttributes = newNameAttributes + a[
0].text + '=' + a[2][0].text
elif a[0].text == 'Add':
newNameAttributes = newNameAttributes + a[
0].text + '=' + a[2][0].text
else:
newNameAttributes = newNameAttributes + a[0].text
for args in a[2]:
newNameAttributes = newNameAttributes + '_' + args.tag + '=' + args.text.replace(
'.', 'p')
newName = str(
fileName.text.replace(
fileExtension, newNameAttributes)) + fileExtension
fileName.text = newName
for folder in augmentedRoot.iter('folder'):
newFolder = str(a[0].text)
folder.text = newFolder
for path in augmentedRoot.iter('path'):
newPath = str(imageDestinationDirectory
) + '/' + a[0].text + '/' + newName
path.text = newPath
for size in augmentedRoot.iter('size'):
size[0].text = str(augmentedImage.shape[1])
size[1].text = str(augmentedImage.shape[0])
size[2].text = str(augmentedImage.shape[2])
i = 0
for obj in augmentedRoot.findall('object'):
if augmentedBBS.bounding_boxes[i].is_fully_within_image(
augmentedImage.shape):
obj[4][0].text = str(augmentedBBS.bounding_boxes[i].x1)
obj[4][1].text = str(augmentedBBS.bounding_boxes[i].y1)
obj[4][2].text = str(augmentedBBS.bounding_boxes[i].x2)
obj[4][3].text = str(augmentedBBS.bounding_boxes[i].y2)
elif augmentedBBS.bounding_boxes[i].is_partly_within_image(
augmentedImage.shape):
augmentedBBS.bounding_boxes[i].clip_out_of_image(
augmentedImage.shape)
obj[4][0].text = str(augmentedBBS.bounding_boxes[i].x1)
obj[4][1].text = str(augmentedBBS.bounding_boxes[i].y1)
obj[4][2].text = str(augmentedBBS.bounding_boxes[i].x2)
obj[4][3].text = str(augmentedBBS.bounding_boxes[i].y2)
else:
augmentedRoot.remove(obj)
i += 1
# draw bounding boxes on image (uncomment to enable)
# augmentedImage = augmentedBBS.draw_on_image(
# augmentedImage, size=5, color=[0, 0, 255])
# saving new xml file
# mkdir if destination directory doesn't exist
if os.path.isdir(newPath.replace(newName, '')) is False:
if os.path.isdir(imageDestinationDirectory) is False:
os.mkdir(imageDestinationDirectory)
os.mkdir(newPath.replace(newName, ''))
# os.mkdir(newPath.replace(newName, '/xmlFile'))
# augmentedTree.write(newPath.replace(newName, '/xmlFile/' + newName).replace(fileExtension, '.xml')) # for saving jpg and xml into seperate directories
if os.path.isdir(
newPath.replace(imageDestinationDirectory,
xmlDestinationDirectory).replace(
newName, '')) is False:
if os.path.isdir(xmlDestinationDirectory) is False:
os.mkdir(xmlDestinationDirectory)
os.mkdir(
newPath.replace(imageDestinationDirectory,
xmlDestinationDirectory).replace(
newName, ''))
augmentedTree.write(
newPath.replace(imageDestinationDirectory,
xmlDestinationDirectory).replace(
fileExtension, '.xml'))
# saving new image
imageio.imsave(newPath, augmentedImage)
# collecting samples to show
if configurationRoot.findall(
'showAugmentedSamples')[0].text == 'Yes':
if finishPartCount is 0 and finishImageCount is 0:
originalImageToShow = image
if finishImageCount is 0: # sample images to show later
augmentedImageToShow.append(augmentedImage)
titlesToShow.append(newNameAttributes[1:])
finishImageCount += 1
print(
str(finishImageCount) + "/" + str(totalImageCount) +
", saved to: " + newPath)
finishPartCount += 1
print('')
# showing augmented samples
# checking whether the user want to show augmented samples
if configurationRoot.findall('showAugmentedSamples')[0].text == 'Yes':
totalSampleCount = math.ceil(augmenterCount / 3)
for i in range(0, augmenterCount, 3):
print('showing sample: ' + str(int((i / 3) + 1)) + '/' +
str(totalSampleCount))
plt.figure(figsize=(12.8, 7.2))
plt.axis('off')
if augmenterCount - i is 1:
plt.subplot(1, 2, 1)
elif augmenterCount - i is 2:
plt.subplot(1, 3, 1)
else:
plt.subplot(2, 2, 1)
plt.imshow(originalImageToShow)
plt.title('Original')
j = 0
while j < 3 and i + j < len(augmentedImageToShow):
if augmenterCount - i is 1:
plt.subplot(1, 2, j + 2)
elif augmenterCount - i is 2:
plt.subplot(1, 3, j + 2)
else:
plt.subplot(2, 2, j + 2)
plt.imshow(augmentedImageToShow[i + j])
plt.title(titlesToShow[i + j])
plt.axis('off')
j += 1
plt.show()
print("Augmentation finished.")
def bbsoi_flipped(self):
# note that x1 and x2 were inverted (otherwise would be x1>x2)
bbs = [ia.BoundingBox(x1=3 - 2, y1=1, x2=3 - 0, y2=3)]
return [ia.BoundingBoxesOnImage(bbs, shape=self.image.shape)]
def getInfo(f, d):
filename_split = os.path.splitext(f)
filename_zero, fileext = filename_split
basename = os.path.basename(filename_zero)
print("file: ", basename)
tree = ET.ElementTree(file=f)
root = tree.getroot()
def chunks(l, n):
for i in range(0, len(l), n):
yield l[i:i+n]
points_list = []
for name in root.findall(".//bndbox/"):
points = [name.text]
#print("points: ", points)
points_list.append(points)
flat_points_list = [int(item) for sublist in points_list for item in sublist]
print("points_list: ", flat_points_list)
img = args.imPath+basename+'.jpg'
img_aug = augment_img(img, d)
img_open = np.array(Image.open(img))
frame_box = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=int(1), y1=int(1), x2=int(img_open.shape[1]-1), y2=int(img_open.shape[0]-1))], shape=img_open.shape)
frame_box = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=int(1), y1=int(1), x2=int(767), y2=int(767))], shape=img_open.shape)
frame_box = ia.KeypointsOnImage([
ia.Keypoint(x=0, y=int(img_open.shape[0])),
ia.Keypoint(x=int(img_open.shape[1]), y=int(img_open.shape[0])),
ia.Keypoint(x=int(img_open.shape[1]), y=0),
ia.Keypoint(x=0, y=0)], shape=img_open.shape)
print("frame_box: ", frame_box)
frame_box_aug = augment_keypoints_frame(frame_box, img_open, d)
print("frame_box_aug: ", frame_box_aug)
frame_box_aug_flat = [int(item) for sublist in frame_box_aug for item in sublist]
print("frame_box_aug_flat: ", frame_box_aug_flat)
frame_box_aug_cv, frame_box_aug_cv_rect = back_forward_convert(frame_box_aug_flat)
print("frame_box_aug_cv: ", frame_box_aug_cv)
img_crop = crop_minAreaRect(img_aug, frame_box_aug_cv_rect)
misc.imsave(args.imPath+basename+'_rotate'+str(d)+'_cropFrame'+'.jpg', img_crop)
pairs = list(zip(*[iter(flat_points_list)] * 2))
chunked = list(chunks(pairs, 4))
print("chunked: ", chunked)
print("pairs: ", pairs)
#img_aug, chunked_points_aug_list = augment(f, chunked)
ordered_points_list = []
bbs_aug_list = []
for chunk in chunked:
flat_points_list_chunk = [item for sublist in chunk for item in sublist]
#flat_points_aug_list_chunk, bbs_aug = augment_keypoints(flat_points_list_chunk, img_aug, d)
flat_points_aug_list_chunk = augment_keypoints(flat_points_list_chunk, img_aug, d)
#bbs_aug_list.append(bbs_aug)
print("flat_points_aug_list_chunk: ", flat_points_aug_list_chunk)
ordered_points_list_chunk = order_points(flat_points_aug_list_chunk)
ordered_points_list.append(ordered_points_list_chunk)
#bbs_aug = bbs_aug_list.on(img_aug)
#img_bbs_aug = bbs_aug.draw_on_image(img_aug, thickness=2)
#misc.imsave(args.imPath+basename+'_rotate'+str(d)+'_crop75_bbsRescaled'+'.jpg', img_bbs_aug)
img_shape = img_aug.shape
print("image shape: ", img_shape)
filename = basename
print("original points list: ", chunked)
print("ordered_points_list: ", ordered_points_list)
filename_aug = filename+'_rotate'+str(d)
#filename_aug = filename
xSorted = sorted(pairs,key=itemgetter(0))
xSorted_extract = [i[0] for i in xSorted]
xmin = int(min(xSorted_extract))
xmax = int(max(xSorted_extract))
ySorted = sorted(pairs,key=itemgetter(1))
ySorted_extract = [i[1] for i in ySorted]
ymin = int(min(ySorted_extract))
ymax = int(min(ySorted_extract))
img_xmin = 0
img_xmax = img_shape[0]
img_ymin = 0
img_ymax = img_shape[1]
#print("xmin, xmax, ymin, ymax: ", xmin, xmax, ymin, ymax)
#print("img_xmin, img_xmax, img_ymin, img_ymax: ", img_xmin, img_xmax, img_ymin, img_ymax)
if xmin < img_xmin:
out_of_bounds.append(filename_aug)
print("box off frame for file: ", filename_aug)
elif xmax > img_xmax:
out_of_bounds.append(filename_aug)
print("box off frame for file: ", filename_aug)
elif ymin < img_ymin:
out_of_bounds.append(filename_aug)
print("box off frame for file: ", filename_aug)
elif ymax > img_ymax:
out_of_bounds.append(filename_aug)
print("box off frame for file: ", filename_aug)
return filename_aug, ordered_points_list, img_shape
def bbsoi_flipped(self):
# note that y1 and y2 were inverted (otherwise would be y1>y2)
bbs = [ia.BoundingBox(x1=0, y1=3 - 3, x2=2, y2=3 - 1)]
return [ia.BoundingBoxesOnImage(bbs, shape=self.image.shape)]
filename = annotations[idx]
height, width, channel = hwc[idx]
label = []
for box in boxes:
label.append(box[0])
bbox_center_x = box[1] * width
bbox_center_y = box[2]* height
bbox_width = box[3] * width
bbox_height = box[4] * height
bbox_top_left_x = bbox_center_x - bbox_width/2
bbox_top_left_y = bbox_center_y - bbox_height/2
bbox_bottom_right_x = bbox_center_x + bbox_width/2
bbox_bottom_right_y = bbox_center_y + bbox_height/2
bbox.append(ia.BoundingBox(bbox_top_left_x, bbox_top_left_y, bbox_bottom_right_x, bbox_bottom_right_y))
bbs = ia.BoundingBoxesOnImage(bbox, shape = image.shape)
# # 만약 어떻게 바운딩박스되는지 보고싶으면
# ia.imshow(bbs.draw_on_image(image, size=2))
seq1 = iaa.Sequential([
iaa.Fliplr(0.5), # horizontal flips
iaa.Crop(percent=(0, 0.1)), # random crops
# Small gaussian blur with random sigma between 0 and 0.5.
# But we only blur about 50% of all images.
iaa.Sometimes(
0.5,
iaa.GaussianBlur(sigma=(0, 0.5)))
])
# Strengthen or weaken the contrast in each image.
seq2 = iaa.LinearContrast((0.75, 1.5))
# Add gaussian noise.
def process_pascal_voc_aug(job):
kwargs = {'Bucket': job.bucket, 'Prefix': job.class_folder}
train_csv_file = job.tempdir + '/train.csv'
test_csv_file = job.tempdir + '/test.csv'
if not os.path.isfile(train_csv_file):
print("creating training csv file")
fout = open(train_csv_file, "w")
fout.write('filename,width,height,class,xmin,ymin,xmax,ymax\n')
fout.close()
if not os.path.isfile(test_csv_file):
print("creating test csv file")
fout = open(test_csv_file, "w")
fout.write('filename,width,height,class,xmin,ymin,xmax,ymax\n')
fout.close()
while True:
resp = job.s3.list_objects_v2(**kwargs)
try:
for obj in resp['Contents']:
_key = obj['Key']
_obj = job.s3.get_object(Bucket=job.bucket, Key=_key)
e = ET.fromstring(_obj['Body'].read().decode('utf-8'))
fname = os.path.basename(e.find('./filename').text)
folder = job.project['id'] + '/train'
bbox = []
params = {'Bucket': job.bucket, 'Key': folder + '/' + fname}
url = job.s3.generate_presigned_url(ClientMethod='get_object',
Params=params)
url_response = http.request('GET', url)
img_array = np.array(bytearray(url_response.data),
dtype=np.uint8)
im = cv2.imdecode(img_array, -1)
for node in e.iter('object'):
xmin = int(node.find('./bndbox/xmin').text)
ymin = int(node.find('./bndbox/ymin').text)
xmax = int(node.find('./bndbox/xmax').text)
ymax = int(node.find('./bndbox/ymax').text)
name = node.find('./name').text
if name not in job.labelmap_dict:
job.labelmap_dict[name] = len(job.labelmap_dict) + 1
bbox.append(
ia.BoundingBox(x1=xmin,
y1=ymin,
x2=xmax,
y2=ymax,
label=node.find('./name').text))
bbs = ia.BoundingBoxesOnImage(bbox, shape=im.shape)
fout = open(train_csv_file, "a")
with concurrent.futures.ThreadPoolExecutor(
max_workers=multiprocessing.cpu_count()) as executor:
future_img = {
executor.submit(create_image, im, job, bbs, fout): i
for i in range(job.train_samples)
}
for future in concurrent.futures.as_completed(future_img):
img = future_img[future]
try:
data = future.result()
except Exception as exc:
print('%r generated an exception: %s' %
(str(i), exc))
fout.close()
fout = open(test_csv_file, "a")
with concurrent.futures.ThreadPoolExecutor(
max_workers=multiprocessing.cpu_count()) as executor:
future_img = {
executor.submit(create_image, im, job, bbs, fout): i
for i in range(job.test_samples)
}
for future in concurrent.futures.as_completed(future_img):
img = future_img[future]
try:
data = future.result()
except Exception as exc:
print('generated an exception: %s' % (exc))
fout.close()
if hasattr(job, 'jb'):
job.jb.meta['current_step_processed'] += 1
job.jb.save_meta()
except KeyError:
break
try:
kwargs['ContinuationToken'] = resp['NextContinuationToken']
except KeyError:
break
job.s3.upload_file(
train_csv_file, job.bucket,
'corpus/' + job.name + "/" + os.path.basename(train_csv_file))
job.s3.upload_file(
test_csv_file, job.bucket,
'corpus/' + job.name + "/" + os.path.basename(test_csv_file))
job.train_csv_file = 'corpus/' + job.name + \
"/" + os.path.basename(train_csv_file)
job.test_csv_file = 'corpus/' + job.name + \
"/" + os.path.basename(test_csv_file)
def img_and_key_point_augmentation(augmentation, img, bbox, key_points):
"""
Augment image and key points, bounding boxes !!
:param augmentation: augmentation settings
:param img: Only one image is needed. Not batch images
:param bbox: [[x1,y1,x2,y2],[x1,y1,x2,y2]...]
:param key_points: [[[x1,y1,x2,y2....],[x1,y1,x2,y2....]...]]
:return: Returns augment image and key points, bbox
"""
# img_copy = img.copy()
image_shape = img.shape
h, w = image_shape[0:2]
# Convert the stochastic sequence of augmenters to a deterministic one.
# The deterministic sequence will always apply the exactly same effects to the images.
det = augmentation.to_deterministic()
ia_bbox = list()
for bounding_box in bbox:
x1, y1, x2, y2 = bounding_box
ia_bbox.append(ia.BoundingBox(x1=x1, y1=y1, x2=x2, y2=y2))
bbs = ia.BoundingBoxesOnImage(ia_bbox, shape=image_shape)
bbs_aug = det.augment_bounding_boxes([bbs])[0]
# img = bbs_aug.draw_on_image(img)
after_bbox = list()
for bounding_box in bbs_aug.bounding_boxes:
bbox_list = [
bounding_box.x1_int, bounding_box.y1_int, bounding_box.x2_int,
bounding_box.y2_int
]
if bbox_list[0] >= w: bbox_list[0] = w - 1
if bbox_list[1] >= h: bbox_list[1] = h - 1
if bbox_list[2] >= w: bbox_list[2] = w - 1
if bbox_list[3] >= h: bbox_list[3] = h - 1
if bbox_list[0] == bbox_list[2] or bbox_list[1] == bbox_list[3]:
return img_and_key_point_augmentation(augmentation, img, bbox,
key_points)
bbox_list = list(map(lambda x: max(x, 0), bbox_list))
after_bbox.append(bbox_list)
after_key_points = list()
for key_point_list in key_points:
after_key_point_list = list()
for key_point in key_point_list:
xy_points = list()
for i, x in enumerate(key_point[::2]):
y = key_point[(i * 2) + 1]
xy_points.append(ia.Keypoint(x=x, y=y))
keypoints_on_image = det.augment_keypoints(
[ia.KeypointsOnImage(xy_points, shape=image_shape)])
# img = keypoints_on_image[0].draw_on_image(img)
xy_points = list()
for key_point in keypoints_on_image[0].keypoints:
kp = [key_point.x_int, key_point.y_int]
if 0 > min(kp) or w <= max(kp[::2]) or h <= max(kp[1::2]):
# print(kp)
return img_and_key_point_augmentation(
augmentation, img, bbox, key_points)
xy_points.extend(kp)
after_key_point_list.append(xy_points)
after_key_points.append(after_key_point_list)
img_aug = det.augment_image(img)
assert img_aug.shape == image_shape, "Augmentation shouldn't change image size"
return img_aug, after_bbox, after_key_points
def main():
parser = argparse.ArgumentParser(description="Check augmenters visually.")
parser.add_argument(
"--only",
default=None,
help=
"If this is set, then only the results of an augmenter with this name will be shown. "
"Optionally, comma-separated list.",
required=False)
args = parser.parse_args()
images = [
ia.quokka_square(size=(128, 128)),
ia.imresize_single_image(data.astronaut(), (128, 128))
]
keypoints = [
ia.KeypointsOnImage([
ia.Keypoint(x=50, y=40, vis=None, label=None),
ia.Keypoint(x=70, y=38, vis=None, label=None),
ia.Keypoint(x=62, y=52, vis=None, label=None)
],
shape=images[0].shape),
ia.KeypointsOnImage([
ia.Keypoint(x=55, y=32, vis=None, label=None),
ia.Keypoint(x=42, y=95, vis=None, label=None),
ia.Keypoint(x=75, y=89, vis=None, label=None)
],
shape=images[1].shape)
]
bounding_boxes = [
ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=10, y1=10, x2=20, y2=20),
ia.BoundingBox(x1=40, y1=50, x2=70, y2=60)
],
shape=images[0].shape),
ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=10, y1=10, x2=20, y2=20),
ia.BoundingBox(x1=40, y1=50, x2=70, y2=60)
],
shape=images[1].shape)
]
augmenters = [
iaa.Sequential([
iaa.CoarseDropout(p=0.5, size_percent=0.05),
iaa.AdditiveGaussianNoise(scale=0.1 * 255),
iaa.Crop(percent=0.1)
],
name="Sequential"),
iaa.SomeOf(2,
children=[
iaa.CoarseDropout(p=0.5, size_percent=0.05),
iaa.AdditiveGaussianNoise(scale=0.1 * 255),
iaa.Crop(percent=0.1)
],
name="SomeOf"),
iaa.OneOf(children=[
iaa.CoarseDropout(p=0.5, size_percent=0.05),
iaa.AdditiveGaussianNoise(scale=0.1 * 255),
iaa.Crop(percent=0.1)
],
name="OneOf"),
iaa.Sometimes(0.5,
iaa.AdditiveGaussianNoise(scale=0.1 * 255),
name="Sometimes"),
iaa.WithColorspace("HSV",
children=[iaa.Add(20)],
name="WithColorspace"),
iaa.WithChannels([0], children=[iaa.Add(20)], name="WithChannels"),
iaa.AddToHueAndSaturation((-20, 20),
per_channel=True,
name="AddToHueAndSaturation"),
iaa.Noop(name="Noop"),
iaa.Resize({
"width": 64,
"height": 64
}, name="Resize"),
iaa.CropAndPad(px=(-8, 8), name="CropAndPad-px"),
iaa.Pad(px=(0, 8), name="Pad-px"),
iaa.Crop(px=(0, 8), name="Crop-px"),
iaa.Crop(percent=(0, 0.1), name="Crop-percent"),
iaa.Fliplr(0.5, name="Fliplr"),
iaa.Flipud(0.5, name="Flipud"),
iaa.Superpixels(p_replace=0.75, n_segments=50, name="Superpixels"),
iaa.Grayscale(0.5, name="Grayscale0.5"),
iaa.Grayscale(1.0, name="Grayscale1.0"),
iaa.GaussianBlur((0, 3.0), name="GaussianBlur"),
iaa.AverageBlur(k=(3, 11), name="AverageBlur"),
iaa.MedianBlur(k=(3, 11), name="MedianBlur"),
iaa.BilateralBlur(d=10, name="BilateralBlur"),
iaa.Sharpen(alpha=(0.1, 1.0), lightness=(0, 2.0), name="Sharpen"),
iaa.Emboss(alpha=(0.1, 1.0), strength=(0, 2.0), name="Emboss"),
iaa.EdgeDetect(alpha=(0.1, 1.0), name="EdgeDetect"),
iaa.DirectedEdgeDetect(alpha=(0.1, 1.0),
direction=(0, 1.0),
name="DirectedEdgeDetect"),
iaa.Add((-50, 50), name="Add"),
iaa.Add((-50, 50), per_channel=True, name="AddPerChannel"),
iaa.AddElementwise((-50, 50), name="AddElementwise"),
iaa.AdditiveGaussianNoise(loc=0,
scale=(0.0, 0.1 * 255),
name="AdditiveGaussianNoise"),
iaa.Multiply((0.5, 1.5), name="Multiply"),
iaa.Multiply((0.5, 1.5), per_channel=True, name="MultiplyPerChannel"),
iaa.MultiplyElementwise((0.5, 1.5), name="MultiplyElementwise"),
iaa.Dropout((0.0, 0.1), name="Dropout"),
iaa.CoarseDropout(p=0.05,
size_percent=(0.05, 0.5),
name="CoarseDropout"),
iaa.Invert(p=0.5, name="Invert"),
iaa.Invert(p=0.5, per_channel=True, name="InvertPerChannel"),
iaa.ContrastNormalization(alpha=(0.5, 2.0),
name="ContrastNormalization"),
iaa.SaltAndPepper(p=0.05, name="SaltAndPepper"),
iaa.Salt(p=0.05, name="Salt"),
iaa.Pepper(p=0.05, name="Pepper"),
iaa.CoarseSaltAndPepper(p=0.05,
size_percent=(0.01, 0.1),
name="CoarseSaltAndPepper"),
iaa.CoarseSalt(p=0.05, size_percent=(0.01, 0.1), name="CoarseSalt"),
iaa.CoarsePepper(p=0.05, size_percent=(0.01, 0.1),
name="CoarsePepper"),
iaa.Affine(scale={
"x": (0.8, 1.2),
"y": (0.8, 1.2)
},
translate_px={
"x": (-16, 16),
"y": (-16, 16)
},
rotate=(-45, 45),
shear=(-16, 16),
order=ia.ALL,
cval=(0, 255),
mode=ia.ALL,
name="Affine"),
iaa.PiecewiseAffine(scale=0.03,
nb_rows=(2, 6),
nb_cols=(2, 6),
name="PiecewiseAffine"),
iaa.PerspectiveTransform(scale=0.1, name="PerspectiveTransform"),
iaa.ElasticTransformation(alpha=(0.5, 8.0),
sigma=1.0,
name="ElasticTransformation"),
iaa.Alpha(factor=(0.0, 1.0),
first=iaa.Add(100),
second=iaa.Dropout(0.5),
per_channel=False,
name="Alpha"),
iaa.Alpha(factor=(0.0, 1.0),
first=iaa.Add(100),
second=iaa.Dropout(0.5),
per_channel=True,
name="AlphaPerChannel"),
iaa.Alpha(factor=(0.0, 1.0),
first=iaa.Affine(rotate=(-45, 45)),
per_channel=True,
name="AlphaAffine"),
iaa.AlphaElementwise(factor=(0.0, 1.0),
first=iaa.Add(50),
second=iaa.ContrastNormalization(2.0),
per_channel=False,
name="AlphaElementwise"),
iaa.AlphaElementwise(factor=(0.0, 1.0),
first=iaa.Add(50),
second=iaa.ContrastNormalization(2.0),
per_channel=True,
name="AlphaElementwisePerChannel"),
iaa.AlphaElementwise(factor=(0.0, 1.0),
first=iaa.Affine(rotate=(-45, 45)),
per_channel=True,
name="AlphaElementwiseAffine"),
iaa.SimplexNoiseAlpha(first=iaa.EdgeDetect(1.0),
per_channel=False,
name="SimplexNoiseAlpha"),
iaa.FrequencyNoiseAlpha(first=iaa.EdgeDetect(1.0),
per_channel=False,
name="FrequencyNoiseAlpha")
]
augmenters.append(
iaa.Sequential([iaa.Sometimes(0.2, aug.copy()) for aug in augmenters],
name="Sequential"))
augmenters.append(
iaa.Sometimes(0.5, [aug.copy() for aug in augmenters],
name="Sometimes"))
for augmenter in augmenters:
if args.only is None or augmenter.name in [
v.strip() for v in args.only.split(",")
]:
print("Augmenter: %s" % (augmenter.name, ))
grid = []
for image, kps, bbs in zip(images, keypoints, bounding_boxes):
aug_det = augmenter.to_deterministic()
imgs_aug = aug_det.augment_images(
np.tile(image[np.newaxis, ...], (16, 1, 1, 1)))
kps_aug = aug_det.augment_keypoints([kps] * 16)
bbs_aug = aug_det.augment_bounding_boxes([bbs] * 16)
imgs_aug_drawn = [
kps_aug_one.draw_on_image(img_aug)
for img_aug, kps_aug_one in zip(imgs_aug, kps_aug)
]
imgs_aug_drawn = [
bbs_aug_one.draw_on_image(img_aug)
for img_aug, bbs_aug_one in zip(imgs_aug_drawn, bbs_aug)
]
grid.append(np.hstack(imgs_aug_drawn))
ia.imshow(np.vstack(grid))
ymax.text = str(new_ymax)
tree.write(os.path.join(root, str(image_id) + "_aug" + '.xml'))
if __name__ == "__main__":
# cmd = os.getcwd()
cmd = "/Users/wangbenkang/Desktop/500/Example"
image_id = "000010"
img = Image.open(os.path.join(cmd, str(image_id) + '.jpg'))
img = np.array(img)
bndbox = read_xml_annotation(cmd, str(image_id) + '.xml')
bbs = ia.BoundingBoxesOnImage(
[
ia.BoundingBox(
x1=bndbox[0], y1=bndbox[1], x2=bndbox[2], y2=bndbox[3]) #改这里！
],
shape=img.shape)
seq = iaa.Sequential([
iaa.Flipud(0.5), # vertically flip 20% of all images
iaa.Multiply((1.2, 1.5)), # change brightness, doesn't affect BBs
iaa.Affine(
translate_px={
"x": 10,
"y": 10
},
scale=(0.8, 0.95),
rotate=(-10, 10)
) # translate by 40/60px on x/y axis, and scale to 50-70%, affects BBs
])
seq_det = seq.to_deterministic() # 保持坐标和图像同步改变，而不是随机
def main(_):
writer = tf.python_io.TFRecordWriter(FLAGS.output_path)
sometimes = lambda aug: iaa.Sometimes(0.5, aug)
seq = iaa.Sequential(
[
iaa.Fliplr(0.5), # horizontally flip 50% of all images
iaa.Flipud(0.2), # vertically flip 20% of all images
# crop images by -10% to 20% of their height/width
sometimes(iaa.CropAndPad(
percent=(-0.05, 0.1),
pad_mode=ia.ALL,
pad_cval=(0, 255)
)),
sometimes(iaa.Affine(
translate_percent={"x": (-0.2, 0.2), "y": (-0.2, 0.2)}, # translate by -50 to +50 percent (per axis)
rotate=(-20, 20), # rotate by -90 to +90 degrees
order=[0, 1], # use nearest neighbour or bilinear interpolation (fast)
cval=(0, 255), # if mode is constant, use a cval between 0 and 255
mode=ia.ALL # use any of scikit-image's warping modes (see 2nd image from the top for examples)
)),
],
random_order=True)
ann = None
num_batches = 80
images, bboxs = [], []
dir_path = '/home/kshitiz/workspace/autonomous_drone_for_tracking/data/detector_' + FLAGS.flag + '/'
xml_files = [f for f in os.listdir(dir_path) if f.endswith('.xml')]
num_orig_samples = len(xml_files)
for i,xml_file in enumerate(xml_files):
path = dir_path + xml_file
tree = ET.parse(path)
root = tree.getroot()
obj = root.findall('object')
bndbox = obj[0].find('bndbox')
xmin = int(bndbox.find('xmin').text)
xmax = int(bndbox.find('xmax').text)
ymin = int(bndbox.find('ymin').text)
ymax = int(bndbox.find('ymax').text)
image = np.asarray(Image.open((path.split('.')[0] + '.png')), dtype='uint8')
bbox = ia.BoundingBoxesOnImage([ia.BoundingBox(x1=xmin, y1=ymin, x2=xmax, y2=ymax)], shape=image.shape)
images.append(image)
bboxs.append(bbox)
# image_copy = image.copy()
# vis_util.draw_bounding_boxes_on_image_array( image_copy,
# np.array([[float(ymin)/image.shape[0], float(xmin)/image.shape[1], float(ymax)/image.shape[0], float(xmax)/image.shape[1] ]]),
# color='yellow',
# thickness=4)
# imgplot = plt.imshow(image_copy)
# plt.show()
for i in range(num_batches):
seq_det = seq.to_deterministic()
aug_images = seq_det.augment_images(images)
aug_bboxs = seq_det.augment_bounding_boxes(bboxs)
for j, (image_np, abbox) in enumerate(zip(aug_images, aug_bboxs)):
#aug_bboxs[]
bbox = abbox.bounding_boxes[0]
result = Image.fromarray(image_np)
out_path = '/home/kshitiz/workspace/autonomous_drone_for_tracking/TF_ObjectDetection/data/' + \
FLAGS.flag + '/' + str(i*num_orig_samples+j).zfill(6) + '.png'
result.save(out_path)
# image = image_np.copy()
# vis_util.draw_bounding_boxes_on_image_array( image,
# np.array([[ float(bbox.y1)/image.shape[0], float(bbox.x1)/image.shape[1], float(bbox.y2)/image.shape[0], float(bbox.x2)/image.shape[1]]]),
# color='yellow',
# thickness=4)
# imgplot = plt.imshow(image)
# plt.show()
tf_example = create_tf_example(out_path, bbox, image.shape)
writer.write(tf_example.SerializeToString())
os.remove(out_path)
if i%10==0 and i!=0:
print "Number of batches processed", i
writer.close()
def test_deepcopy(self):
batch = ia.Batch()
observed = batch.deepcopy()
keys = list(observed.__dict__.keys())
assert len(keys) >= 12
for attr_name in keys:
assert getattr(observed, attr_name) is None
batch = ia.Batch(images=np.zeros((1, 1, 3), dtype=np.uint8))
observed = batch.deepcopy()
for attr_name in observed.__dict__.keys():
if attr_name != "images_unaug":
assert getattr(observed, attr_name) is None
assert ia.is_np_array(observed.images_unaug)
batch = ia.Batch(
images=np.zeros((1, 1, 3), dtype=np.uint8),
heatmaps=[
ia.HeatmapsOnImage(np.zeros((1, 1, 1), dtype=np.float32),
shape=(4, 4, 3))
],
segmentation_maps=[
ia.SegmentationMapOnImage(np.zeros((1, 1), dtype=np.int32),
shape=(5, 5, 3),
nb_classes=20)
],
keypoints=[
ia.KeypointsOnImage([ia.Keypoint(x=1, y=2)], shape=(6, 6, 3))
],
bounding_boxes=[
ia.BoundingBoxesOnImage(
[ia.BoundingBox(x1=1, y1=2, x2=3, y2=4)], shape=(7, 7, 3))
],
polygons=[
ia.PolygonsOnImage([ia.Polygon([(0, 0), (10, 0), (10, 10)])],
shape=(100, 100, 3))
],
data={
"test": 123,
"foo": "bar",
"test2": [1, 2, 3]
})
observed = batch.deepcopy()
for attr_name in observed.__dict__.keys():
if "_unaug" not in attr_name and attr_name != "data":
assert getattr(observed, attr_name) is None
assert ia.is_np_array(observed.images_unaug)
assert observed.images_unaug.shape == (1, 1, 3)
assert isinstance(observed.heatmaps_unaug[0], ia.HeatmapsOnImage)
assert isinstance(observed.segmentation_maps_unaug[0],
ia.SegmentationMapOnImage)
assert isinstance(observed.keypoints_unaug[0], ia.KeypointsOnImage)
assert isinstance(observed.bounding_boxes_unaug[0],
ia.BoundingBoxesOnImage)
assert isinstance(observed.polygons_unaug[0], ia.PolygonsOnImage)
assert isinstance(observed.data, dict)
assert observed.heatmaps_unaug[0].shape == (4, 4, 3)
assert observed.segmentation_maps_unaug[0].shape == (5, 5, 3)
assert observed.keypoints_unaug[0].shape == (6, 6, 3)
assert observed.bounding_boxes_unaug[0].shape == (7, 7, 3)
assert observed.polygons_unaug[0].shape == (100, 100, 3)
assert observed.heatmaps_unaug[0].arr_0to1.shape == (1, 1, 1)
assert observed.segmentation_maps_unaug[0].arr.shape == (1, 1, 20)
assert observed.keypoints_unaug[0].keypoints[0].x == 1
assert observed.keypoints_unaug[0].keypoints[0].y == 2
assert observed.bounding_boxes_unaug[0].bounding_boxes[0].x1 == 1
assert observed.bounding_boxes_unaug[0].bounding_boxes[0].y1 == 2
assert observed.bounding_boxes_unaug[0].bounding_boxes[0].x2 == 3
assert observed.bounding_boxes_unaug[0].bounding_boxes[0].y2 == 4
assert observed.polygons_unaug[0].polygons[0].exterior[0, 0] == 0
assert observed.polygons_unaug[0].polygons[0].exterior[0, 1] == 0
assert observed.polygons_unaug[0].polygons[0].exterior[1, 0] == 10
assert observed.polygons_unaug[0].polygons[0].exterior[1, 1] == 0
assert observed.polygons_unaug[0].polygons[0].exterior[2, 0] == 10
assert observed.polygons_unaug[0].polygons[0].exterior[2, 1] == 10
assert observed.data["test"] == 123
assert observed.data["foo"] == "bar"
assert observed.data["test2"] == [1, 2, 3]
def test_BoundingBoxesOnImage():
reseed()
# test height/width
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=45, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
assert bbsoi.height == 40
assert bbsoi.width == 50
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=45, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2],
shape=np.zeros((40, 50, 3),
dtype=np.uint8))
assert bbsoi.height == 40
assert bbsoi.width == 50
# empty
bb = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb], shape=(40, 50, 3))
assert not bbsoi.empty
bbsoi = ia.BoundingBoxesOnImage([], shape=(40, 50, 3))
assert bbsoi.empty
# on()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=45, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2],
shape=np.zeros((40, 50, 3),
dtype=np.uint8))
bbsoi_projected = bbsoi.on((40, 50))
assert bbsoi_projected.bounding_boxes[0].y1 == 10
assert bbsoi_projected.bounding_boxes[0].x1 == 20
assert bbsoi_projected.bounding_boxes[0].y2 == 30
assert bbsoi_projected.bounding_boxes[0].x2 == 40
assert bbsoi_projected.bounding_boxes[1].y1 == 15
assert bbsoi_projected.bounding_boxes[1].x1 == 25
assert bbsoi_projected.bounding_boxes[1].y2 == 35
assert bbsoi_projected.bounding_boxes[1].x2 == 45
bbsoi_projected = bbsoi.on((40 * 2, 50 * 2, 3))
assert bbsoi_projected.bounding_boxes[0].y1 == 10 * 2
assert bbsoi_projected.bounding_boxes[0].x1 == 20 * 2
assert bbsoi_projected.bounding_boxes[0].y2 == 30 * 2
assert bbsoi_projected.bounding_boxes[0].x2 == 40 * 2
assert bbsoi_projected.bounding_boxes[1].y1 == 15 * 2
assert bbsoi_projected.bounding_boxes[1].x1 == 25 * 2
assert bbsoi_projected.bounding_boxes[1].y2 == 35 * 2
assert bbsoi_projected.bounding_boxes[1].x2 == 45 * 2
bbsoi_projected = bbsoi.on(np.zeros((40 * 2, 50 * 2, 3), dtype=np.uint8))
assert bbsoi_projected.bounding_boxes[0].y1 == 10 * 2
assert bbsoi_projected.bounding_boxes[0].x1 == 20 * 2
assert bbsoi_projected.bounding_boxes[0].y2 == 30 * 2
assert bbsoi_projected.bounding_boxes[0].x2 == 40 * 2
assert bbsoi_projected.bounding_boxes[1].y1 == 15 * 2
assert bbsoi_projected.bounding_boxes[1].x1 == 25 * 2
assert bbsoi_projected.bounding_boxes[1].y2 == 35 * 2
assert bbsoi_projected.bounding_boxes[1].x2 == 45 * 2
# from_xyxy_array()
bbsoi = ia.BoundingBoxesOnImage.from_xyxy_array(np.float32(
[[0.0, 0.0, 1.0, 1.0], [1.0, 2.0, 3.0, 4.0]]),
shape=(40, 50, 3))
assert len(bbsoi.bounding_boxes) == 2
assert np.allclose(bbsoi.bounding_boxes[0].x1, 0.0)
assert np.allclose(bbsoi.bounding_boxes[0].y1, 0.0)
assert np.allclose(bbsoi.bounding_boxes[0].x2, 1.0)
assert np.allclose(bbsoi.bounding_boxes[0].y2, 1.0)
assert np.allclose(bbsoi.bounding_boxes[1].x1, 1.0)
assert np.allclose(bbsoi.bounding_boxes[1].y1, 2.0)
assert np.allclose(bbsoi.bounding_boxes[1].x2, 3.0)
assert np.allclose(bbsoi.bounding_boxes[1].y2, 4.0)
assert bbsoi.shape == (40, 50, 3)
bbsoi = ia.BoundingBoxesOnImage.from_xyxy_array(np.int32([[0, 0, 1, 1],
[1, 2, 3, 4]]),
shape=(40, 50, 3))
assert len(bbsoi.bounding_boxes) == 2
assert np.allclose(bbsoi.bounding_boxes[0].x1, 0.0)
assert np.allclose(bbsoi.bounding_boxes[0].y1, 0.0)
assert np.allclose(bbsoi.bounding_boxes[0].x2, 1.0)
assert np.allclose(bbsoi.bounding_boxes[0].y2, 1.0)
assert np.allclose(bbsoi.bounding_boxes[1].x1, 1.0)
assert np.allclose(bbsoi.bounding_boxes[1].y1, 2.0)
assert np.allclose(bbsoi.bounding_boxes[1].x2, 3.0)
assert np.allclose(bbsoi.bounding_boxes[1].y2, 4.0)
assert bbsoi.shape == (40, 50, 3)
bbsoi = ia.BoundingBoxesOnImage.from_xyxy_array(np.zeros((0, 4),
dtype=np.float32),
shape=(40, 50, 3))
assert len(bbsoi.bounding_boxes) == 0
assert bbsoi.shape == (40, 50, 3)
# to_xyxy_array()
xyxy_arr = np.float32([[0.0, 0.0, 1.0, 1.0], [1.0, 2.0, 3.0, 4.0]])
bbsoi = ia.BoundingBoxesOnImage.from_xyxy_array(xyxy_arr,
shape=(40, 50, 3))
xyxy_arr_out = bbsoi.to_xyxy_array()
assert np.allclose(xyxy_arr, xyxy_arr_out)
assert xyxy_arr_out.dtype == np.float32
xyxy_arr_out = bbsoi.to_xyxy_array(dtype=np.int32)
assert np.allclose(xyxy_arr.astype(np.int32), xyxy_arr_out)
assert xyxy_arr_out.dtype == np.int32
xyxy_arr_out = ia.BoundingBoxesOnImage(
[], shape=(40, 50, 3)).to_xyxy_array(dtype=np.int32)
assert xyxy_arr_out.shape == (0, 4)
# draw_on_image()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=45, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
image = bbsoi.draw_on_image(np.zeros(bbsoi.shape, dtype=np.uint8),
color=[0, 255, 0],
alpha=1.0,
thickness=1,
copy=True,
raise_if_out_of_image=False)
assert np.all(image[10 - 1, 20 - 1, :] == [0, 0, 0])
assert np.all(image[10 - 1, 20 - 0, :] == [0, 0, 0])
assert np.all(image[10 - 0, 20 - 1, :] == [0, 0, 0])
assert np.all(image[10 - 0, 20 - 0, :] == [0, 255, 0])
assert np.all(image[10 + 1, 20 + 1, :] == [0, 0, 0])
assert np.all(image[30 - 1, 40 - 1, :] == [0, 0, 0])
assert np.all(image[30 + 1, 40 - 0, :] == [0, 0, 0])
assert np.all(image[30 + 0, 40 + 1, :] == [0, 0, 0])
assert np.all(image[30 + 0, 40 + 0, :] == [0, 255, 0])
assert np.all(image[30 + 1, 40 + 1, :] == [0, 0, 0])
assert np.all(image[15 - 1, 25 - 1, :] == [0, 0, 0])
assert np.all(image[15 - 1, 25 - 0, :] == [0, 0, 0])
assert np.all(image[15 - 0, 25 - 1, :] == [0, 0, 0])
assert np.all(image[15 - 0, 25 - 0, :] == [0, 255, 0])
assert np.all(image[15 + 1, 25 + 1, :] == [0, 0, 0])
assert np.all(image[35 - 1, 45 - 1, :] == [0, 0, 0])
assert np.all(image[35 + 1, 45 + 0, :] == [0, 0, 0])
assert np.all(image[35 + 0, 45 + 1, :] == [0, 0, 0])
assert np.all(image[35 + 0, 45 + 0, :] == [0, 255, 0])
assert np.all(image[35 + 1, 45 + 1, :] == [0, 0, 0])
# remove_out_of_image()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
bbsoi_slim = bbsoi.remove_out_of_image(fully=True, partly=True)
assert len(bbsoi_slim.bounding_boxes) == 1
assert bbsoi_slim.bounding_boxes[0] == bb1
# clip_out_of_image()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
eps = np.finfo(np.float32).eps
bbsoi_clip = bbsoi.clip_out_of_image()
assert len(bbsoi_clip.bounding_boxes) == 2
assert bbsoi_clip.bounding_boxes[0].y1 == 10
assert bbsoi_clip.bounding_boxes[0].x1 == 20
assert bbsoi_clip.bounding_boxes[0].y2 == 30
assert bbsoi_clip.bounding_boxes[0].x2 == 40
assert bbsoi_clip.bounding_boxes[1].y1 == 15
assert bbsoi_clip.bounding_boxes[1].x1 == 25
assert bbsoi_clip.bounding_boxes[1].y2 == 35
assert 50 - 2 * eps < bbsoi_clip.bounding_boxes[1].x2 < 50
# shift()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
bbsoi_shifted = bbsoi.shift(right=1)
assert len(bbsoi_shifted.bounding_boxes) == 2
assert bbsoi_shifted.bounding_boxes[0].y1 == 10
assert bbsoi_shifted.bounding_boxes[0].x1 == 20 - 1
assert bbsoi_shifted.bounding_boxes[0].y2 == 30
assert bbsoi_shifted.bounding_boxes[0].x2 == 40 - 1
assert bbsoi_shifted.bounding_boxes[1].y1 == 15
assert bbsoi_shifted.bounding_boxes[1].x1 == 25 - 1
assert bbsoi_shifted.bounding_boxes[1].y2 == 35
assert bbsoi_shifted.bounding_boxes[1].x2 == 51 - 1
# copy()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
bbsoi_copy = bbsoi.copy()
assert len(bbsoi.bounding_boxes) == 2
assert bbsoi_copy.bounding_boxes[0].y1 == 10
assert bbsoi_copy.bounding_boxes[0].x1 == 20
assert bbsoi_copy.bounding_boxes[0].y2 == 30
assert bbsoi_copy.bounding_boxes[0].x2 == 40
assert bbsoi_copy.bounding_boxes[1].y1 == 15
assert bbsoi_copy.bounding_boxes[1].x1 == 25
assert bbsoi_copy.bounding_boxes[1].y2 == 35
assert bbsoi_copy.bounding_boxes[1].x2 == 51
bbsoi.bounding_boxes[0].y1 = 0
assert bbsoi_copy.bounding_boxes[0].y1 == 0
# deepcopy()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
bbsoi_copy = bbsoi.deepcopy()
assert len(bbsoi.bounding_boxes) == 2
assert bbsoi_copy.bounding_boxes[0].y1 == 10
assert bbsoi_copy.bounding_boxes[0].x1 == 20
assert bbsoi_copy.bounding_boxes[0].y2 == 30
assert bbsoi_copy.bounding_boxes[0].x2 == 40
assert bbsoi_copy.bounding_boxes[1].y1 == 15
assert bbsoi_copy.bounding_boxes[1].x1 == 25
assert bbsoi_copy.bounding_boxes[1].y2 == 35
assert bbsoi_copy.bounding_boxes[1].x2 == 51
bbsoi.bounding_boxes[0].y1 = 0
assert bbsoi_copy.bounding_boxes[0].y1 == 10
# repr() / str()
bb1 = ia.BoundingBox(y1=10, x1=20, y2=30, x2=40, label=None)
bb2 = ia.BoundingBox(y1=15, x1=25, y2=35, x2=51, label=None)
bbsoi = ia.BoundingBoxesOnImage([bb1, bb2], shape=(40, 50, 3))
bb1_expected = "BoundingBox(x1=20.0000, y1=10.0000, x2=40.0000, y2=30.0000, label=None)"
bb2_expected = "BoundingBox(x1=25.0000, y1=15.0000, x2=51.0000, y2=35.0000, label=None)"
expected = "BoundingBoxesOnImage([%s, %s], shape=(40, 50, 3))" % (
bb1_expected, bb2_expected)
assert bbsoi.__repr__() == bbsoi.__str__() == expected
def _ndarray_to_BoundingBoxesOnImage(bboxes: np.ndarray, img_shape) -> ia.BoundingBoxesOnImage:
bb_list = []
for bbox in bboxes:
bb_list.append(ia.BoundingBox(*bbox))
return ia.BoundingBoxesOnImage(bb_list, shape=img_shape)
