def imgDistort(filepath: str, flag=True):
if isinstance(filepath, str):
if os.path.exists(filepath):
img = io.imread(filepath)
else:
raise FileNotFoundError('Image file not found!')
elif isinstance(filepath, np.ndarray):
img = filepath
else:
logger.error('Input file error')
return
imgShape = img.shape
imgWidth = imgShape[1]
imgHeight = imgShape[0]
center = (random.randint(int(0.2 * imgWidth), int(0.8 * imgWidth)),
random.randint(int(0.2 * imgHeight), int(0.8 * imgHeight)))
mouse = (random.randint(int(0.2 * imgWidth), int(0.8 * imgWidth)),
random.randint(int(0.2 * imgHeight), int(0.8 * imgHeight)))
radious = random.randint(50, 250)
input_img = Image.fromarray(img).convert('RGB')
resImg = imgFilter(input_img, radious, center, mouse)
resImg = np.array(resImg)
if flag:
image_path, _ = os.path.splitext(filepath)
io.imsave(image_path + '_distort.jpg', resImg)
logger.info('Saved to {}'.format(image_path + '_distort.jpg'))
else:
return resImg
def imgAug_withLabels(imgPath, labelPath, number=1, yamlFilePath=''):
"""
number : file number you want to generate.
"""
logger.info("currently, only *.jpg supported")
oriImgs = glob.glob(imgPath + os.sep +
'*.jpg') if os.path.isdir(imgPath) else [imgPath]
jsonFiles = glob.glob(labelPath + os.sep +
'*.json') if os.path.isdir(labelPath) else [
labelPath
]
if type(number) != int:
logger.error('Augumentation times error.Using 1 as default')
number = 1
else:
if number < 1:
logger.warning(
'Augumentation times {} is less than 1.Using 1 as default'.
format(number))
number = 1
for num in range(0, number):
for i in tqdm(oriImgs):
i_json = i.replace(
imgPath, labelPath) if os.path.isdir(labelPath) else labelPath
i_json = i_json.replace('.jpg', '.json')
imgAug.aug_labelme(i, i_json, num=num, yamlFilePath=yamlFilePath)
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
if self.rect_or_poly == 'rect':
return crop.multiRectanleCrop(self.img, self.cropNumber,
self.noise)
elif self.rect_or_poly == 'poly':
return crop.multiPolygonCrop(self.img, self.cropNumber,
self.noise, self.convexHull)
else:
res = []
for i in self.img:
if self.rect_or_poly == 'rect':
res.append(
crop.multiRectanleCrop(i, self.cropNumber,
self.noise))
elif self.rect_or_poly == 'poly':
res.append(
crop.multiPolygonCrop(i, self.cropNumber,
self.noise, self.convexHull))
return res
else:
logger.error('Images are not found!')
def do(self):
if len(self.augs) == 0:
logger.info('None methods founds')
return
pimgs = self.imgs
# print(pimgs)
for i in self.operations:
i.setImgs(pimgs)
pimgs = i.do()
gc.collect()
if not self.saveFile:
return pimgs
else:
if not os.path.exists(self.saveDir) and self.saveDir != '':
os.mkdir(self.saveDir)
else:
logger.error("Provided savedir is not valid")
return
if type(self.imgs) is not list:
if type(self.imgs) is str:
filename = os.path.split(self.imgs)[1]
else:
filename = 'test.jpg'
io.imsave(self.saveDir + os.sep + filename, pimgs)
else:
for i in range(len(self.imgs)):
if isinstance(self.augs[i], str):
filename = os.path.split(i)[1]
else:
filename = 'test{}.jpg'.format(i)
io.imsave(self.saveDir + os.sep + filename, pimgs[i])
logger.info("Done! See {}.".format(self.saveDir))
def readYamlFile(yamlFile):
if isinstance(yamlFile, str) and os.path.exists(yamlFile):
if yamlFile.endswith('.yaml'):
f = open(yamlFile, 'r', encoding='utf-8')
x = yaml.load(f.read(), Loader=yaml.FullLoader)
try:
res = x['label_names']
return res
except:
return None
finally:
f.close()
elif yamlFile.endswith('.txt'):
with open(yamlFile, 'r', encoding='utf-8') as f:
classList = f.readlines()
res = labels2yaml(classList, savefile=False)
return res['label_names']
elif isinstance(yamlFile, dict):
try:
res = x['label_names']
return res
except:
return None
else:
logger.error(
'input type error. must be a .txt file or .yaml file or a dict like \{"label_names":["classA":1,...,"classN":10]\}'
)
def labels2yaml(labels: list, savePath='', savefile=True):
# pass
if '_background_' not in labels:
labels.append('_background_')
tmp = dict()
tmp['_background_'] = 0
classId = 1
for i in range(0, len(labels)):
x = labels[i].replace('\n','').strip()
if x != '_background_':
tmp[x] = classId
classId += 1
data = dict()
data['label_names'] = tmp
del tmp
if savefile:
try:
with open(savePath + os.sep + 'info.yaml', 'w',
encoding='utf-8') as f:
yaml.dump(data, f, allow_unicode=True)
logger.info("successfully convert labels to yaml, see {}".format(
savePath + os.sep + 'info.yaml'))
except Exception as e:
logger.error(e)
return data
def imgAug_LabelImg(imgPath, xmlpath, number=1):
"""
number : file number you want to generate.
"""
logger.info("currently, only *.jpg supported")
oriImgs = glob.glob(imgPath + os.sep +
'*.jpg') if os.path.isdir(imgPath) else [imgPath]
xmlFiles = glob.glob(xmlpath + os.sep +
'*.xml') if os.path.isdir(xmlpath) else [xmlpath]
if type(number) != int:
logger.error('Augumentation times error.Using 1 as default')
number = 1
else:
if number < 1:
logger.warning(
'Augumentation times {} is less than 1.Using 1 as default'.
format(number))
number = 1
pool = Pool(__CPUS__ - 1)
pool_list = []
for i in oriImgs:
for num in tqdm(range(0, number)):
resultspool = pool.apply_async(proc_xml,
(i, imgPath, xmlpath, num))
pool_list.append(resultspool)
logger.info('successfully create {} tasks'.format(len(pool_list)))
for pr in tqdm(pool_list):
re_list = pr.get()
def getJsons(imgPath, maskPath, savePath, yamlPath=''):
"""
imgPath: origin image path \n
maskPath : mask image path \n
savePath : json file save path \n
>>> getJsons(path-to-your-imgs,path-to-your-maskimgs,path-to-your-jsonfiles)
"""
logger.info("currently, only *.jpg supported")
if os.path.isfile(imgPath):
getMultiShapes.getMultiShapes(imgPath, maskPath, savePath, yamlPath)
elif os.path.isdir(imgPath):
oriImgs = glob.glob(imgPath + os.sep + '*.jpg')
maskImgs = glob.glob(maskPath + os.sep + '*.jpg')
for i in tqdm(oriImgs):
i_mask = i.replace(imgPath, maskPath)
if os.path.exists(i_mask):
# print(i)
getMultiShapes.getMultiShapes(i, i_mask, savePath, yamlPath)
else:
logger.warning('corresponding mask image not found!')
continue
else:
logger.error(
'input error. got [{},{},{},{}]. file maybe missing.'.format(
imgPath, maskPath, savePath, yamlPath))
logger.info('Done! See here. {}'.format(savePath))
def load(stream, Loader=None):
if isinstance(stream, str):
lis = readYamlFile(stream)
elif isinstance(stream, _io.TextIOWrapper):
lis = stream.readlines()
else:
logger.error('input parameter error.')
return
res = dict()
sections, secNameNum, secNum = getSection(lis)
def readYamlFile(filepath: str, encoding='utf-8'):
if not os.path.exists(filepath):
logger.error('file not found')
return
with open(filepath, 'r', encoding=encoding) as f:
# print(type(f))
lis = f.readlines()
# print(lis)
return lis
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return distort.imgDistort(self.img, flag=False)
else:
res = []
for i in self.img:
res.append(distort.imgDistort(i, flag=False))
return res
else:
logger.error('Images are not found!')
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return imgZoom(self.img, self.size, False).get('zoom').oriImg
else:
res = []
for i in self.img:
res.append(imgZoom(i, 1.0, False).get('zoom').oriImg)
return res
else:
logger.error('Images are not found!')
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return perspective.persTrans(self.img, self.factor)
else:
res = []
for i in self.img:
res.append(perspective.persTrans(i, self.factor))
return res
else:
logger.error('Images are not found!')
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return imgNoise(self.img, False,
self.noiseType).get('noise').oriImg
else:
res = []
for i in self.img:
res.append(imgNoise(i, False, []).get('noise').oriImg)
return res
else:
logger.error('Images are not found!')
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return imgTranslation(self.img, False, self.th,
self.tv).get('trans').oriImg
else:
res = []
for i in self.img:
res.append(
imgTranslation(i, False, 0, 0).get('trans').oriImg)
return res
else:
logger.error('Images are not found!')
def go(self, **kwargs):
if not os.path.exists(self.img_or_path) or not os.path.exists(
self.label_or_path):
logger.error('File or path not exists!')
return
# single file
if os.path.isfile(self.label_or_path):
# xml format
_, filename = os.path.split(self.label_or_path)
if self.saveFilePath == '':
savePath = self.label_or_path
else:
savePath = self.saveFilePath + os.sep + filename
go_single(self.label_or_path, self.img_or_path,
self.negativeNumbers, self.iou, self.negativeClassName,
savePath)
else:
xmls = glob.glob(self.label_or_path + os.sep + '*.xml')
if not self.multiProcesses:
for i in tqdm(xmls):
_, filename = os.path.split(i)
if self.saveFilePath == '':
savePath = self.label_or_path
else:
savePath = self.saveFilePath + os.sep + filename
imgpath = self.img_or_path + os.sep + filename.replace(
'.xml', '.jpg')
go_single(i, imgpath, self.negativeNumbers, self.iou,
self.negativeClassName, savePath)
else:
pool = Pool(__CPUS__ - 1)
pool_list = []
for i in xmls:
_, filename = os.path.split(i)
if self.saveFilePath == '':
savePath = self.label_or_path
else:
savePath = self.saveFilePath + os.sep + filename
imgpath = self.img_or_path + os.sep + filename.replace(
'.xml', '.jpg')
resultspool = pool.apply_async(
go_single, (i, imgpath, self.negativeNumbers, self.iou,
self.negativeClassName, savePath))
pool_list.append(resultspool)
for pr in tqdm(pool_list):
re_list = pr.get()
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
return imgRotation(self.img, self.angle, self.scale,
False).get('rotation').oriImg
else:
res = []
for i in self.img:
res.append(
imgRotation(i, self.angle, self.scale,
False).get('rotation').oriImg)
return res
else:
logger.error('Images are not found!')
def imgZoom(oriImg, size, flag=True):
"""
size : The zoom factor along the axes, default 0.8~1.8
"""
# pass
if isinstance(oriImg, str):
if os.path.exists(oriImg):
img = io.imread(oriImg)
else:
raise FileNotFoundError('Original image not found')
elif isinstance(oriImg, np.ndarray):
img = oriImg
else:
logger.error('input {} type error'.format('oriImg'))
return
try:
size = float(size)
# if size == 0.0:
# raise ValueError('zoom factor cannot be zero')
except:
logger.warning('input {} type error ,got {}.'.format(
'size', type(size)))
size = random.uniform(0.8, 1.8)
size = round(size, 2)
if size <= 0 or size == 1:
size = random.uniform(0.8, 1.8)
size = round(size, 2)
resOri = _getZoomedImg(img, size)
if flag:
parent_path = os.path.dirname(oriImg)
if os.path.exists(parent_path + os.sep + 'augimgs_'):
pass
else:
os.makedirs(parent_path + os.sep + 'augimgs_')
tmp = os.path.splitext(oriImg)[0]
fileName = tmp.split(os.sep)[-1]
io.imsave(
parent_path + os.sep + 'augimgs_' + os.sep + fileName +
'_zoom.jpg', resOri)
else:
d = dict()
d['zoom'] = Ori_Pro(resOri, None)
return d
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
if self.rect_or_poly == 'rect':
return inpaint.rectangleInpaint(self.img, self.startPoint)
elif self.rect_or_poly == 'poly':
return inpaint.polygonInpaint(self.img, self.startPoint)
else:
res = []
for i in self.img:
res.append(inpaint.polygonInpaint(i, self.startPoint))
return res
else:
logger.error('Images are not found!')
def getPolygonPascal(xmlPath):
in_file = open(xmlPath)
tree = ET.parse(in_file)
root = tree.getroot()
shapes = []
try:
for obj in root.iter('object'):
flags = {}
group_id = 'null'
shape_type = 'polygon'
# pass
dic = dict()
label = obj.find('name').text
polygon = obj.find('bndbox')
# print(len(polygon))
# if len(polygon)>2:
# points = []
# for i in range(0,len(polygon)):
xmin = int(polygon.find('xmin').text)
ymin = int(polygon.find('ymin').text)
xmax = int(polygon.find('xmax').text)
ymax = int(polygon.find('ymax').text)
# print(polygon.find('point{}'.format(i)).text)
# tmp = polygon.find('point{}'.format(i)).text.split(',')
# point = [int(tmp[0]),int(tmp[1])]
# print(point)
p1 = [xmin, ymin]
p2 = [xmax, ymax]
p3 = [xmin, ymax]
p4 = [xmax, ymin]
# points.append(point)
points = [p1, p2, p3, p4]
# del tmp,point
dic['flags'] = flags
dic['group_id'] = group_id
dic['shape_type'] = shape_type
dic['points'] = points
dic['label'] = label
shapes.append(dic)
# print(shapes)
return shapes
except Exception:
logger.error(traceback.print_exc())
def imgAug_withoutLabels(imgPath, number=1):
logger.info("currently, only *.jpg supported")
oriImgs = glob.glob(imgPath + os.sep +
'*.jpg') if os.path.isdir(imgPath) else [imgPath]
if type(number) != int:
logger.error('Augumentation times error.Using 1 as default')
number = 1
else:
if number < 1:
logger.warning(
'Augumentation times {} is less than 1.Using 1 as default'.
format(number))
number = 1
for num in range(0, number):
for i in tqdm(oriImgs):
imgAug_nolabel.aug(i, num=num)
def mosiac_img(imgs: list, heightFactor=0.5, widthFactor=0.5):
if not type(imgs) is list:
logger.error('Input must be a list!')
return
if len(imgs) == 0:
logger.error('None image found!')
return
if len(imgs) == 1:
for _ in range(0, 3):
imgs.append(imgs[0])
if len(imgs) == 2:
for _ in range(0, 2):
imgs.append(imgs[0])
if len(imgs) == 3:
for _ in range(0, 1):
imgs.append(imgs[0])
mHeight, mWidth = getMeanSize(imgs)
img_left_top = resize_img(
np.array(skimage.transform.resize(imgs[0], (mHeight, mWidth)) *
255).astype(np.uint8), heightFactor, widthFactor)
img_right_top = resize_img(
np.array(skimage.transform.resize(imgs[1], (mHeight, mWidth)) *
255).astype(np.uint8), heightFactor, 1 - widthFactor)
img_left_bottom = resize_img(
np.array(skimage.transform.resize(imgs[2], (mHeight, mWidth)) *
255).astype(np.uint8), 1 - heightFactor, widthFactor)
img_right_bottom = resize_img(
np.array(skimage.transform.resize(imgs[3], (mHeight, mWidth)) *
255).astype(np.uint8), 1 - heightFactor, 1 - widthFactor)
h1 = np.hstack((img_left_top, img_right_top))
h2 = np.hstack((img_left_bottom, img_right_bottom))
return np.vstack((h1, h2))
def do(self):
if self.img is not None:
if not self.getXmls:
if isinstance(self.img, str) or isinstance(
self.img, np.ndarray):
return resize.resize_img(self.img, self.heightFactor,
self.widthFactor)
else:
res = []
for i in self.img:
res.append(
resize.resize_img(i, self.heightFactor,
self.widthFactor))
return res
else:
return resize.resizeScript(self.img, self.xml,
self.heightFactor, self.widthFactor)
else:
logger.error('Images are not found!')
def split(folder: str, savaFolder: str, nc: int = 42, multiprocesses=True):
global RES
txts = glob.glob(folder + os.sep + "*.txt")
if not len(txts) > 0:
logger.error('Folder is empty, none txt file found!')
return
trainFile = open(savaFolder + os.sep + 'train.txt', 'w+', encoding='utf-8')
trainvalFile = open(savaFolder + os.sep + 'val.txt',
'w+',
encoding='utf-8')
logger.info('======== start analysing ========')
pool = Pool(__CPUS__ - 1)
pool_list = []
for t in txts:
dic = dict()
for k in range(0, nc):
dic[k] = 0
resultpool = pool.apply_async(getInfo, (t, dic))
pool_list.append(resultpool)
for pr in tqdm(pool_list):
res = pr.get()
RES.append(res)
tmp = getSum(RES)
r = getMean(tmp, nc)
getTrainValSet(RES, r)
global valList
for i in list(set(valList)):
trainvalFile.write(i.filename + '\n')
trainList = list(set(RES).difference(set(valList)))
for i in trainList:
trainFile.write(i.filename + '\n')
trainvalFile.close()
trainFile.close()
def do(self):
if self.img is not None:
if isinstance(self.img, str) or isinstance(self.img, np.ndarray):
r = imgFlip(self.img, flag=False)
imgH = r.get('h').oriImg
imgV = r.get('v').oriImg
imgHV = r.get('h_v').oriImg
return [imgH, imgV, imgHV]
else:
res = []
for i in self.img:
r = imgFlip(i, flag=False)
imgH = r.get('h').oriImg
imgV = r.get('v').oriImg
imgHV = r.get('h_v').oriImg
res.append([imgH, imgV, imgHV])
return res
else:
logger.error('Images are not found!')
def getXmls(imgPath, maskPath, savePath):
logger.info("currently, only *.jpg supported")
if os.path.isfile(imgPath):
getMultiShapes.getMultiObjs_voc(imgPath, maskPath, savePath)
elif os.path.isdir(imgPath):
oriImgs = glob.glob(imgPath + os.sep + '*.jpg')
maskImgs = glob.glob(maskPath + os.sep + '*.jpg')
for i in tqdm(oriImgs):
i_mask = i.replace(imgPath, maskPath)
# print(i)
if os.path.exists(i_mask):
getMultiShapes.getMultiObjs_voc(i, i_mask, savePath)
else:
logger.warning('corresponding mask image not found!')
continue
else:
logger.error('input error. got [{},{},{}]. file maybe missing.'.format(
imgPath, maskPath, savePath))
logger.info('Done! See here. {}'.format(savePath))
def generateMaskFile(data, yamlFile):
imageHeight = data["imageHeight"]
imageWidth = data["imageWidth"]
mask = np.zeros((imageHeight, imageWidth)).astype(np.uint8)
yamlInfomation = readYamlFile(yamlFile)
# print(yamlInfomation)
try:
shapes = data['shapes']
# print(shapes)
for i in shapes:
k = i["label"]
# print(k)
v = readYamlFileVals(yamlInfomation, k)
# print(v)
polygon = i["points"]
if v != 0:
# print('lllllllllllllllllll')
mask = draw_mask(polygon, mask, int(v))
# print("<==================>".format(np.max(mask)))
return mask
except Exception as e:
logger.error(e)
def getPolygon(xmlPath):
in_file = open(xmlPath)
tree = ET.parse(in_file)
root = tree.getroot()
shapes = []
try:
for obj in root.iter('object'):
flags = {}
group_id = 'null'
shape_type = 'polygon'
# pass
dic = dict()
label = obj.find('name').text
polygon = obj.find('polygon')
# print(len(polygon))
if len(polygon) > 2:
points = []
for i in range(0, len(polygon)):
# print(polygon.find('point{}'.format(i)).text)
tmp = polygon.find('point{}'.format(i)).text.split(',')
point = [int(tmp[0]), int(tmp[1])]
# print(point)
points.append(point)
del tmp, point
dic['flags'] = flags
dic['group_id'] = group_id
dic['shape_type'] = shape_type
dic['points'] = points
dic['label'] = label
shapes.append(dic)
# print(shapes)
return shapes
except Exception:
logger.error(traceback.print_exc())
def x2jConvert_pascal(xmlpath, originImgPath, flag=True):
# pass
if not os.path.exists(xmlpath) or not os.path.exists(originImgPath):
logger.error('file not exist')
return
base64Code = imgEncode(originImgPath)
shapes = getPolygonPascal(xmlpath)
(fatherPath, filename_ext) = os.path.split(originImgPath)
(filename, _) = os.path.splitext(filename_ext)
ob = dict()
ob['imageData'] = base64Code
ob['flags'] = {}
ob['version'] = labelmeVersion
ob['imagePath'] = filename_ext
img = io.imread(originImgPath)
imgShape = img.shape
del img
ob['imageHeight'] = imgShape[0]
ob['imageWidth'] = imgShape[1]
ob['shapes'] = shapes
if flag:
with open(fatherPath + os.sep + filename + '_p.json',
'w',
encoding='utf-8') as f:
j = json.dumps(ob, sort_keys=True, indent=4)
f.write(j)
logger.info('save to path {}'.format(fatherPath + os.sep + filename +
'_p.json'))
return fatherPath + os.sep + filename + '_p.json'
else:
return json.dumps(ob, sort_keys=True, indent=4)
def mosiacScript_no_reshape(imgs: list, xmls: list, savePath: str, flag=False):
heightFactor = random.uniform(0.1, 0.5)
widthFactor = random.uniform(0.1, 0.5)
img1, img2, img3, img4 = imgs[0], imgs[1], imgs[2], imgs[3]
if not type(imgs) is list or not type(xmls) is list:
logger.error('Input must be list!')
return
imgname = getName(xmls)
# imgname = 'test1123'
folder = savePath
mosiacImg, res, _, _ = mosiac_img_no_reshape(imgs, heightFactor,
widthFactor)
front = res[0]
# print(front)
# print(heightFactor_,widthFactor_)
heightFactor = min(heightFactor, 1 - heightFactor)
widthFactor = min(widthFactor, 1 - widthFactor)
# print(heightFactor, widthFactor)
tree1 = ET.parse(xmls[0])
tree2 = resizeScript(img2,
xmls[1],
heightFactor=img1.shape[0] / img2.shape[0],
widthFactor=img1.shape[1] / img2.shape[1],
flag=False)
tree3 = resizeScript(img3,
xmls[2],
heightFactor=img1.shape[0] / img3.shape[0],
widthFactor=img1.shape[1] / img3.shape[1],
flag=False)
tree4 = resizeScript(img4,
xmls[3],
heightFactor=img1.shape[0] / img4.shape[0],
widthFactor=img1.shape[1] / img4.shape[1],
flag=False)
root1 = tree1.getroot()
root2 = tree2.getroot()
for box in root2.iter('bndbox'):
xmin = float(box.find('xmin').text)
ymin = float(box.find('ymin').text)
xmax = float(box.find('xmax').text)
ymax = float(box.find('ymax').text)
box.find('xmin').text = str(
int(xmin + widthFactor * mosiacImg.shape[1]))
box.find('xmax').text = str(
int(xmax + widthFactor * mosiacImg.shape[1]))
root3 = tree3.getroot()
for box in root3.iter('bndbox'):
xmin = float(box.find('xmin').text)
ymin = float(box.find('ymin').text)
xmax = float(box.find('xmax').text)
ymax = float(box.find('ymax').text)
box.find('ymin').text = str(
int(ymin + heightFactor * mosiacImg.shape[0]))
box.find('ymax').text = str(
int(ymax + heightFactor * mosiacImg.shape[0]))
root4 = tree4.getroot()
for box in root4.iter('bndbox'):
xmin = float(box.find('xmin').text)
ymin = float(box.find('ymin').text)
xmax = float(box.find('xmax').text)
ymax = float(box.find('ymax').text)
box.find('xmin').text = str(
int(xmin + widthFactor * mosiacImg.shape[1]))
box.find('xmax').text = str(
int(xmax + widthFactor * mosiacImg.shape[1]))
box.find('ymin').text = str(
int(ymin + heightFactor * mosiacImg.shape[0]))
box.find('ymax').text = str(
int(ymax + heightFactor * mosiacImg.shape[0]))
boxes = []
r1, r2, r3, r4 = getBoxes(front,
root1,
root2,
root3,
root4,
mosiacImg.shape,
heightFactor=heightFactor,
widthFactor=widthFactor)
for box in r1.iter('object'):
# print(box)
boxes.append(box)
for box in r2.iter('object'):
# print(box)
boxes.append(box)
for box in r3.iter('object'):
# print(box)
boxes.append(box)
for box in r4.iter('object'):
# print(box)
boxes.append(box)
# print(len(boxes))
imgshape = mosiacImg.shape
objs = []
for o in boxes:
obj = dict()
name = o.find('name').text
difficult = 0
xmlbox = o.find('bndbox')
b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text),
float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text))
tmp = dict()
tmp['xmin'] = str(int(b[0]))
tmp['ymin'] = str(int(b[2]))
tmp['xmax'] = str(int(b[1]))
tmp['ymax'] = str(int(b[3]))
# print(tmp)
if not (int(tmp['xmin']) == 0 and int(tmp['xmax']) == 0
and int(tmp['ymin']) == 0 and int(tmp['ymax']) == 0):
obj['name'] = name
obj['difficult'] = difficult
obj['bndbox'] = tmp
objs.append(obj)
del tmp
# print(len(objs))
tmpPath = savePath + os.sep + imgname + '.xml'
filepath = tmpPath.replace('.xml', '.jpg')
filename = imgname + '.jpg'
img2xml_multiobj(tmpPath, tmpPath, folder, filename, filepath, imgshape[1],
imgshape[0], objs)
logger.info('Saved to {}.'.format(tmpPath))
if flag:
skimage.io.imsave(filepath, mosiacImg)