1 前言
目标检测框标注格式目前常用的有3种，分别是json（createML格式），xml(VOC格式)，txt(yolo格式),这里用python实现3种格式之间的互相转换，并且最后提供了python打包的exe小工具。
2 实现方式
2.1 解码json（createML格式）标注文件
下面是解码json格式的函数，读取json文件，将json文件的标注信息转换为一个字典形式，便于后面的转换工作：
def get_json_info(src_path, src_image_path, classes_name):
# print(src_path)
json_info = json.load(open(src_path, "r", encoding="utf-8"))
for i, bbox in enumerate(json_info["shapes"]):
if bbox['label'] not in classes_name:
del json_info["shapes"][i]
return json_info
2.2 解码xml(VOC格式)标注文件
下面是解码xml格式的函数，读取xml文件，将xml文件的标注信息转换为一个字典形式，便于后面的转换工作：
def get_xml_info(src_path, src_image_path, classes_name):
in_file = open(src_path, "r", encoding='utf-8')
tree = ET.parse(in_file)
root = tree.getroot()
xml_info = dict()
xml_info["shapes"] = list()
xml_info["imagePath"] = root.find("filename").text
size = root.find("size")
if int(size.find("height").text) == 0 or int(size.find("width").text) == 0:
img = cv2.imdecode(np.fromfile(src_image_path, dtype=np.uint8), -1)
# cv2.imencode('.jpg', img)[1].tofile(path)
xml_info["imageHeight"], xml_info["imageWidth"], _ = img.shape
else:
xml_info["imageHeight"] = int(size.find("height").text)
xml_info["imageWidth"] = int(size.find("width").text)
if int(xml_info["imageHeight"]) == 0 or int(xml_info["imageWidth"]) == 0:
print(src_path)
for obj in root.iter("object"):
if obj.find('name').text not in classes_name:
# if obj.find('name').text not in classes_name.keys():
continue
bbox_shape = dict()
bbox_shape["label"] = obj.find('name').text
# bbox_shape["label"] =classes_name[obj.find('name').text]
bndbox = obj.find("bndbox")
xmin = float(bndbox.find("xmin").text)
xmax = float(bndbox.find("xmax").text)
ymin = float(bndbox.find("ymin").text)
ymax = float(bndbox.find("ymax").text)
bbox_shape["points"] = [[xmax, ymax], [xmin, ymin]]
xml_info["shapes"].append(bbox_shape)
return xml_info
2.3 解码txt(yolo格式)标注文件
下面是解码txt格式的函数，读取txt文件，将txt文件的标注信息转换为一个字典形式，便于后面的转换工作：
def get_txt_info(src_path, src_image_path, classes_name):
with open(src_path, "r", encoding="utf-8") as f:
lines = list(map(lambda x: x[:-1].split(" "), f.readlines()))
txt_info = dict()
txt_info["shapes"] = list()
txt_info["imagePath"] = os.path.split(src_image_path)[-1]
if os.path.isfile(src_image_path):
img = cv2.imdecode(np.fromfile(src_image_path, dtype=np.uint8), -1)
# cv2.imencode('.' + os.path.split(src_path)[-1], img)[1].tofile(tar_path)
txt_info["imageHeight"], txt_info["imageWidth"], _ = img.shape
else:
print("图片{}不存在，需要根据图片获取图片高宽".format(src_image_path))
# exit()
for line in lines:
bbox_shape = dict()
bbox_shape["label"] = classes_name[int(line[0])]
# print(line)
cx = float(line[1]) * txt_info["imageWidth"]
cy = float(line[2]) * txt_info["imageHeight"]
w = float(line[3]) * txt_info["imageWidth"]
h = float(line[4]) * txt_info["imageHeight"]
xmin = cx - w / 2
xmax = cx + w / 2
ymin = cy - h / 2
ymax = cy + h / 2
# print([xmax, xmin, ymax, ymin])
bbox_shape["points"] = [[xmax, ymax], [xmin, ymin]]
txt_info["shapes"].append(bbox_shape)
return txt_info
2.4 编码json（createML格式）标注文件
下面是编码json格式的函数，将上述解码成字典形式的标注信息编码成json（createML格式）标注文件。
def to_json(target_path, info,classes_name=[]):
json_info = dict()
json_info["version"] = "4.5.10"
json_info["flags"] = dict()
json_info["shapes"] = list()
for bbox in info["shapes"]:
json_bbox = dict()
json_bbox["label"] = bbox["label"]
json_bbox["points"] = bbox["points"]
json_bbox["group_id"] = None
json_bbox["shape_type"] = "rectangle"
json_bbox["flags"] = dict()
json_info["shapes"].append(json_bbox)
json_info["imagePath"] = info["imagePath"]
json_info["imageData"] = None
json_info["imageHeight"] = info["imageHeight"]
json_info["imageWidth"] = info["imageWidth"]
# print(json_info)
with open(target_path, "w") as f:
f.write(json.dumps(json_info, indent=2, separators=(',', ': ')))
return
2.5 编码xml(VOC格式)标注文件
下面是编码xml格式的函数，将上述解码成字典形式的标注信息编码成xml(VOC格式)标注文件。
def to_xml(target_path, info, classes_name=[]):
annotation = ET.Element("annotation")
tree = ET.ElementTree(annotation)
folder = ET.SubElement(annotation, 'folder')
folder.text = 'images'
filename = ET.SubElement(annotation, 'filename')
filename.text = info["imagePath"]
size = ET.SubElement(annotation, 'size')
width = ET.SubElement(size, "width")
width.text = str(info["imageWidth"])
height = ET.SubElement(size, "height")
height.text = str(info["imageHeight"])
depth = ET.SubElement(size, "depth")
depth.text = str(3)
for bbox_shape in info["shapes"]: # 有多个框
points = bbox_shape["points"]
points = np.array(points)
label = bbox_shape["label"]
# print(label)
xmin_value = min(points[:, 0])
xmax_value = max(points[:, 0])
ymin_value = min(points[:, 1])
ymax_value = max(points[:, 1])
# print(xmin_value)
if xmax_value <= xmin_value:
pass
elif ymax_value <= ymin_value:
pass
else:
object = ET.SubElement(annotation, "object")
name = ET.SubElement(object, "name")
name.text = label
pose = ET.SubElement(object, "pose")
pose.text = "Unspecified"
truncated = ET.SubElement(object, "truncated")
truncated.text = str(0)
difficult = ET.SubElement(object, "difficult")
difficult.text = str(0)
bndbox = ET.SubElement(object, "bndbox")
xmin = ET.SubElement(bndbox, "xmin")
xmin.text = str(int(xmin_value))
xmax = ET.SubElement(bndbox, "xmax")
xmax.text = str(int(xmax_value))
ymin = ET.SubElement(bndbox, "ymin")
ymin.text = str(int(ymin_value))
ymax = ET.SubElement(bndbox, "ymax")
ymax.text = str(int(ymax_value))
pretty_xml(annotation, '\t', '\n')
tree.write(target_path, encoding="utf-8", xml_declaration=True)
return
def pretty_xml(element, indent, newline, level=0): #
if element:
if (element.text is None) or element.text.isspace():
element.text = newline + indent * (level + 1)
else:
element.text = newline + indent * (level + 1) + element.text.strip() + newline + indent * (level + 1)
temp = list(element)
for subelement in temp:
if temp.index(subelement) < (len(temp) - 1):
subelement.tail = newline + indent * (level + 1)
else:
subelement.tail = newline + indent * level
pretty_xml(subelement, indent, newline, level=level + 1)
2.6 编码txt(yolo格式)标注文件
下面是编码txt格式的函数，将上述解码成字典形式的标注信息编码成txt(yolo格式)标注文件。
def to_txt(target_path, info, classes_name=[]):
if classes_name == []:
print("classes_name不能为空，需要传入classes_name参数")
return
with open(target_path, "w") as f:
for bbox in info["shapes"]:
class_id = str(classes_name.index(bbox["label"]))
points = np.array(bbox["points"])
xmin = min(points[:, 0])
xmax = max(points[:, 0])
ymin = min(points[:, 1])
ymax = max(points[:, 1])
dw = 1.0 / info["imageWidth"]
dh = 1.0 / info["imageHeight"]
cx = (xmax + xmin) / 2.0
cy = (ymax + ymin) / 2.0
w = xmax - xmin
h = ymax - ymin
cx = str(cx * dw)
cy = str(cy * dh)
w = str(w * dw)
h = str(h * dh)
f.write(class_id + " " + cx + " " + cy + " " + w + " " + h + "\n")
3. 总代码
上述编码解码文件写好后，剩下的就是使用这些函数，完成转换了，下面就使用这些函数的一个总代码。
import os
import math
import numpy as np
import cv2
import shutil
import json
from xml.etree import ElementTree as ET
def pretty_xml(element, indent, newline, level=0): #
if element:
if (element.text is None) or element.text.isspace():
element.text = newline + indent * (level + 1)
else:
element.text = newline + indent * (level + 1) + element.text.strip() + newline + indent * (level + 1)
temp = list(element)
for subelement in temp:
if temp.index(subelement) < (len(temp) - 1):
subelement.tail = newline + indent * (level + 1)
else:
subelement.tail = newline + indent * level
pretty_xml(subelement, indent, newline, level=level + 1)
def get_json_info(src_path, src_image_path, classes_name):
# print(src_path)
json_info = json.load(open(src_path, "r", encoding="utf-8"))
for i, bbox in enumerate(json_info["shapes"]):
if bbox['label'] not in classes_name:
del json_info["shapes"][i]
return json_info
def get_txt_info(src_path, src_image_path, classes_name):
with open(src_path, "r", encoding="utf-8") as f:
lines = list(map(lambda x: x[:-1].split(" "), f.readlines()))
txt_info = dict()
txt_info["shapes"] = list()
txt_info["imagePath"] = os.path.split(src_image_path)[-1]
if os.path.isfile(src_image_path):
img = cv2.imdecode(np.fromfile(src_image_path, dtype=np.uint8), -1)
# cv2.imencode('.' + os.path.split(src_path)[-1], img)[1].tofile(tar_path)
txt_info["imageHeight"], txt_info["imageWidth"], _ = img.shape
else:
print("图片{}不存在，需要根据图片获取图片高宽".format(src_image_path))
# exit()
for line in lines:
bbox_shape = dict()
bbox_shape["label"] = classes_name[int(line[0])]
# print(line)
cx = float(line[1]) * txt_info["imageWidth"]
cy = float(line[2]) * txt_info["imageHeight"]
w = float(line[3]) * txt_info["imageWidth"]
h = float(line[4]) * txt_info["imageHeight"]
xmin = cx - w / 2
xmax = cx + w / 2
ymin = cy - h / 2
ymax = cy + h / 2
# print([xmax, xmin, ymax, ymin])
bbox_shape["points"] = [[xmax, ymax], [xmin, ymin]]
txt_info["shapes"].append(bbox_shape)
return txt_info
def get_xml_info(src_path, src_image_path, classes_name):
in_file = open(src_path, "r", encoding='utf-8')
tree = ET.parse(in_file)
root = tree.getroot()
xml_info = dict()
xml_info["shapes"] = list()
xml_info["imagePath"] = root.find("filename").text
size = root.find("size")
if int(size.find("height").text) == 0 or int(size.find("width").text) == 0:
img = cv2.imdecode(np.fromfile(src_image_path, dtype=np.uint8), -1)
# cv2.imencode('.jpg', img)[1].tofile(path)
xml_info["imageHeight"], xml_info["imageWidth"], _ = img.shape
else:
xml_info["imageHeight"] = int(size.find("height").text)
xml_info["imageWidth"] = int(size.find("width").text)
if int(xml_info["imageHeight"]) == 0 or int(xml_info["imageWidth"]) == 0:
print(src_path)
for obj in root.iter("object"):
if obj.find('name').text not in classes_name:
# if obj.find('name').text not in classes_name.keys():
continue
bbox_shape = dict()
bbox_shape["label"] = obj.find('name').text
# bbox_shape["label"] =classes_name[obj.find('name').text]
bndbox = obj.find("bndbox")
xmin = float(bndbox.find("xmin").text)
xmax = float(bndbox.find("xmax").text)
ymin = float(bndbox.find("ymin").text)
ymax = float(bndbox.find("ymax").text)
bbox_shape["points"] = [[xmax, ymax], [xmin, ymin]]
xml_info["shapes"].append(bbox_shape)
return xml_info
def to_json(target_path, info,classes_name=[]):
json_info = dict()
json_info["version"] = "4.5.10"
json_info["flags"] = dict()
json_info["shapes"] = list()
for bbox in info["shapes"]:
json_bbox = dict()
json_bbox["label"] = bbox["label"]
json_bbox["points"] = bbox["points"]
json_bbox["group_id"] = None
json_bbox["shape_type"] = "rectangle"
json_bbox["flags"] = dict()
json_info["shapes"].append(json_bbox)
json_info["imagePath"] = info["imagePath"]
json_info["imageData"] = None
json_info["imageHeight"] = info["imageHeight"]
json_info["imageWidth"] = info["imageWidth"]
# print(json_info)
with open(target_path, "w") as f:
f.write(json.dumps(json_info, indent=2, separators=(',', ': ')))
def to_xml(target_path, info, classes_name=[]):
annotation = ET.Element("annotation")
tree = ET.ElementTree(annotation)
folder = ET.SubElement(annotation, 'folder')
folder.text = 'images'
filename = ET.SubElement(annotation, 'filename')
filename.text = info["imagePath"]
size = ET.SubElement(annotation, 'size')
width = ET.SubElement(size, "width")
width.text = str(info["imageWidth"])
height = ET.SubElement(size, "height")
height.text = str(info["imageHeight"])
depth = ET.SubElement(size, "depth")
depth.text = str(3)
for bbox_shape in info["shapes"]: # 有多个框
points = bbox_shape["points"]
points = np.array(points)
label = bbox_shape["label"]
# print(label)
xmin_value = min(points[:, 0])
xmax_value = max(points[:, 0])
ymin_value = min(points[:, 1])
ymax_value = max(points[:, 1])
# print(xmin_value)
if xmax_value <= xmin_value:
pass
elif ymax_value <= ymin_value:
pass
else:
object = ET.SubElement(annotation, "object")
name = ET.SubElement(object, "name")
name.text = label
pose = ET.SubElement(object, "pose")
pose.text = "Unspecified"
truncated = ET.SubElement(object, "truncated")
truncated.text = str(0)
difficult = ET.SubElement(object, "difficult")
difficult.text = str(0)
bndbox = ET.SubElement(object, "bndbox")
xmin = ET.SubElement(bndbox, "xmin")
xmin.text = str(int(xmin_value))
xmax = ET.SubElement(bndbox, "xmax")
xmax.text = str(int(xmax_value))
ymin = ET.SubElement(bndbox, "ymin")
ymin.text = str(int(ymin_value))
ymax = ET.SubElement(bndbox, "ymax")
ymax.text = str(int(ymax_value))
pretty_xml(annotation, '\t', '\n')
tree.write(target_path, encoding="utf-8", xml_declaration=True)
return
def to_txt(target_path, info, classes_name=[]):
if classes_name == []:
print("classes_name不能为空，需要传入classes_name参数")
return
with open(target_path, "w") as f:
for bbox in info["shapes"]:
class_id = str(classes_name.index(bbox["label"]))
points = np.array(bbox["points"])
xmin = min(points[:, 0])
xmax = max(points[:, 0])
ymin = min(points[:, 1])
ymax = max(points[:, 1])
dw = 1.0 / info["imageWidth"]
dh = 1.0 / info["imageHeight"]
cx = (xmax + xmin) / 2.0
cy = (ymax + ymin) / 2.0
w = xmax - xmin
h = ymax - ymin
cx = str(cx * dw)
cy = str(cy * dh)
w = str(w * dw)
h = str(h * dh)
f.write(class_id + " " + cx + " " + cy + " " + w + " " + h + "\n")
def makedir(path):
if not os.path.exists(path):
os.makedirs(path)
if __name__ == "__main__":
classes_name = ["smoke", "wire"]
classes_name.append(classes_name)
src = r""
tar = src
src_label_format = "txt"
tar_label_format = "xml"
is_copyimg = False
is_getnolabelimg =False
for phase in ["train", "val"]:
tar_labels_dir = os.path.join(tar, "labels_" + tar_label_format, phase)
makedir(tar_labels_dir)
if is_copyimg:
tar_images_dir = os.path.join(tar, "images", phase)
makedir(tar_images_dir)
if is_getnolabelimg:
noLableDir = os.path.join(tar, "noLabelDir", phase)
makedir(noLableDir)
for image_name in os.listdir(os.path.join(src, "images", phase)):
# print(image_name)
image_path = os.path.join(src, "images", phase, image_name)
src_label_name = os.path.splitext(image_name)[0] + "." + src_label_format
tar_label_name = os.path.splitext(image_name)[0] + "." + tar_label_format
src_label_path = os.path.join(src, "labels_" + src_label_format, phase, src_label_name)
print(src_label_path)
if not os.path.exists(src_label_path):
info = []
try:
info = eval("get_{}_info".format(src_label_format))(src_label_path, image_path, classes_name)
print(info)
except:
info = []
if info != [] and info["shapes"] == []:
info = []
print(info)
if info != []:
l = [shape["label"] for shape in info["shapes"]]
if "del" in l:
shutil.move(image_path, "./res")
shutil.move(src_label_path, "./res_label")
continue
if info:
tar_label_path = os.path.join(tar, "labels_" + tar_label_format, phase, tar_label_name)
eval("to_{}".format(tar_label_format))(tar_label_path, info, classes_name)
if is_copyimg:
if is_getnolabelimg:
if info:
shutil.copy(image_path, tar_images_dir)
else:
shutil.copy(image_path, noLableDir)
else:
shutil.copy(image_path, tar_images_dir)
用python打包好的exe软件
最后提供一个打包好的装换工具，避免需要配置环境
扫一扫
5413
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
