def crop_xml(crop_img_name,
xml_tree,
top_left_x,
top_left_y,
crop_w,
crop_h,
iou_thr=0.5):
'''xml目标框裁剪变换
Args:
crop_img_name:裁剪图片命名
xml_tree：待crop的xml ET.parse()
top_left_x,top_left_y: 裁剪图像左上角坐标
crop_w,crop_h: 裁剪图像宽高
iou_thr: iou阈值
return:
createdxml : 创建的xml CreateXML对象
'''
root = xml_tree.getroot()
size = root.find('size')
depth = int(size.find('depth').text)
createdxml = CreateXML(crop_img_name, int(crop_w), int(crop_h), depth)
for obj in root.iter('object'):
obj_name = obj.find('name').text
xml_box = obj.find('bndbox')
xmin = float(xml_box.find('xmin').text)
ymin = float(xml_box.find('ymin').text)
xmax = float(xml_box.find('xmax').text)
ymax = float(xml_box.find('ymax').text)
box = crop_box([xmin, ymin, xmax, ymax], top_left_x, top_left_y,
crop_w, crop_h, iou_thr)
if (box[0] >= box[2]) or (box[1] >= box[3]):
continue
createdxml.add_object_node(obj_name, box[0], box[1], box[2], box[3])
return createdxml
def rot_xml(rot_img_name,xml_tree,cterxy,rot_angle,scale=1.0,correction=True):
'''
旋转xml文件
Args:
xml_tree: 待旋转xml ET.parse()
cterxy: 旋转中心坐标[cter_x,cter_y]
rot_img_name: 旋转后图片保存名字
rot_angle：旋转角度
scale:放缩尺度
correction: bool,修正旋转后的目标框为正常左上右下坐标
return:
createdxml : 创建的xml CreateXML对象
'''
root = xml_tree.getroot()
size = root.find('size')
imgw,imgh,depth = int(size.find('width').text),int(size.find('height').text),int(size.find('depth').text)
createdxml = CreateXML(rot_img_name,imgw,imgh,depth)
for obj in root.iter('object'):
obj_name = obj.find('name').text
xml_box = obj.find('bndbox')
xmin = int(xml_box.find('xmin').text)
ymin = int(xml_box.find('ymin').text)
xmax = int(xml_box.find('xmax').text)
ymax = int(xml_box.find('ymax').text)
#边框坐标[x1,y1,x2,y2,x3,y3,x4,y4]，左上开始，逆时针
box=rot_box([xmin,ymin,xmax,ymax],cterxy,[imgw,imgh],rot_angle,scale,correction)
rxmin,rymin,rxmax,rymax = utils.confine(box[0],0,imgw-1),utils.confine(box[1],0,imgh-1),utils.confine(box[4],0,imgw-1),utils.confine(box[5],0,imgh-1)
if (rxmin >= rxmax) or (rymin >= rymax):
continue
createdxml.add_object_node(obj_name,box[0],box[1],box[4],box[5])
return createdxml
def flip_xml(flip_img_name,xml_tree,flip_type):
'''翻转xml
Args:
flip_img_name:翻转后图片保存名
xml_tree:待翻转的xml ET.parse()
flip_type:翻转类型,1水平翻转，0垂直翻转，-1水平垂直翻转
return:
createdxml : 创建的xml CreateXML对象
'''
root = xml_tree.getroot()
size = root.find('size')
imgw,imgh,depth = int(size.find('width').text),int(size.find('height').text),int(size.find('depth').text)
createdxml = CreateXML(flip_img_name,int(imgw),int(imgh),depth)
for obj in root.iter('object'):
obj_name = obj.find('name').text
xml_box = obj.find('bndbox')
xmin = int(xml_box.find('xmin').text)
ymin = int(xml_box.find('ymin').text)
xmax = int(xml_box.find('xmax').text)
ymax = int(xml_box.find('ymax').text)
box = flip_box([xmin,ymin,xmax,ymax],imgw,imgh,flip_type)
if (box[0] >= box[2]) or (box[1] >= box[3]):
continue
createdxml.add_object_node(obj_name,box[0],box[1],box[2],box[3])
return createdxml
def generate_img_xml(img_save_name, imgw, imgh, part_imgw, part_imgh,
transforms, imgs_dir, xmls_dir):
'''生成拼接图和拼接xml
Args:
img_save_name:
imgw,imgh:生成总图宽高
transforms:转换操作
imgs_dir:图源目录
xmls_dir:图源对应的xml目录
return:
img:总图
createdxml：总图创建的xml,ET.parse()
'''
createdxml = CreateXML(img_save_name, imgw, imgh, 3)
img = np.zeros((imgh, imgw, 3), dtype=np.uint8)
part_cols = int(imgw / part_imgw)
part_rows = int(imgh / part_imgh)
for row in range(part_rows):
for col in range(part_cols):
start_row = row * part_imgh
start_col = col * part_imgw
part_img_file = utils.randomChoiceIn(imgs_dir)
part_img = cv2.imread(os.path.join(imgs_dir, part_img_file))
part_xml_file = os.path.join(xmls_dir,
part_img_file.split('.')[0] + '.xml')
part_xml_tree = voc_xml.get_xml_tree(part_xml_file)
img, createdxml = mosaic_img_xml(img, part_img, createdxml,
part_xml_tree, transforms,
start_row, start_col)
return img, createdxml
def resize_xml(resized_img_name, xml_tree, dsize=(0, 0), fx=1.0, fy=1.0):
'''
xml目标框放缩变换
Args:
resized_img_name: resize图片保存名
xml_tree: 待resize xml ET.parse()
dsize:指定放缩大小(w,h)
fx,fy:比例放缩
return:
createdxml : 创建的xml CreateXML对象
'''
root = xml_tree.getroot()
size = root.find('size')
imgw, imgh, depth = int(size.find('width').text), int(
size.find('height').text), int(size.find('depth').text)
resize_imgw, resize_imgh = imgw, imgh
print("原始尺寸:", imgw, imgh)
if dsize == (0, 0):
resize_imgw = int(imgw * fx)
resize_imgh = int(imgh * fy)
else:
resize_imgw, resize_imgh = dsize
print("缩放尺寸:", resize_imgw, resize_imgh)
rsize_fx, resize_fy = resize_imgw / imgw, resize_imgh / imgh
createdxml = CreateXML(resized_img_name, resize_imgw, resize_imgh, depth)
for obj in root.iter('object'):
obj_name = obj.find('name').text
xml_box = obj.find('bndbox')
xmin = int(xml_box.find('xmin').text)
ymin = int(xml_box.find('ymin').text)
xmax = int(xml_box.find('xmax').text)
ymax = int(xml_box.find('ymax').text)
box = resize_box([xmin, ymin, xmax, ymax], rsize_fx, resize_fy)
if (box[0] >= box[2]) or (box[1] >= box[3]):
continue
createdxml.add_object_node(obj_name, box[0], box[1], box[2], box[3])
return createdxml
def transform_img_xml(src_imgpath,src_xmlpath,transforms,img_save_name):
'''按transforms中的转换操作变换img和xml
Args:
src_imgpath: 待变换的图片路径
src_xmlpath: xml标注文件路径
transforms：转换操作
img_save_name: 图片保存名
return:
transformed_img:转换完成的图片
createdxml:转换生成的新标签
'''
src_img = cv2.imread(src_imgpath)
src_xml = voc_xml.get_xml_tree(src_xmlpath)
transformed_img,certain_transforms = mosaic.transform_img(src_img,transforms)
imgh,imgw,n_channels = transformed_img.shape
createdxml = CreateXML(img_save_name,imgw,imgh,n_channels)
createdxml = mosaic.transform_xml(src_xml,createdxml,certain_transforms,0,0)
return transformed_img,createdxml