键错误：从Pascal VOC训练tflite文件时"姿势"

Question

我使用tflite_model_maker包来训练object_detector tflite模型。当我尝试从pascal voc导入数据时，我得到KeyError: 'pose'错误。我做错了什么?

Traceback (most recent call last):
  File "2main.py", line 7, in <module>
    dataloader = object_detector.DataLoader.from_pascal_voc('data/images', 'data/annotations', label_map=label_map)
  File "/home/luke/moneyDetector/venv/lib/python3.8/site-packages/tensorflow_examples/lite/model_maker/core/data_util/object_detector_dataloader.py", line 217, in from_pascal_voc
    cache_writer.write_files(
  File "/home/luke/moneyDetector/venv/lib/python3.8/site-packages/tensorflow_examples/lite/model_maker/core/data_util/object_detector_dataloader_util.py", line 252, in write_files
    tf_example = create_pascal_tfrecord.dict_to_tf_example(
  File "/home/luke/moneyDetector/venv/lib/python3.8/site-packages/tensorflow_examples/lite/model_maker/third_party/efficientdet/dataset/create_pascal_tfrecord.py", line 176, in dict_to_tf_example
    poses.append(obj['pose'].encode('utf8'))
KeyError: 'pose'

编辑:对于进一步的上下文，我认为注释是通过从cvat 导出创建的。

Answer 1

这是因为你的注释文件没有属性'pose'。当我使用CVAT作为注释工具并意识到"姿势"one_answers"截断"缺失时，我也遇到了同样的问题。因此，我将我的注释数据导入到另一个注释工具(LabelImg)，然后再次导出到pascal voc。我不认为这是最好的方法，但它有效。

Answer 2

当我从CVAT工具导出数据时，我也遇到了同样的问题，我已经设法用这个脚本修复了它:

import xmltodict
from dict2xml import dict2xml
import glob

def postprocessor(path, key, value):
    if key == 'object':
        if 'pose' not in value:
            value['pose'] = 'Unspecified'
        value['bndbox']['xmax'] = int(float(value['bndbox']['xmax']))
        value['bndbox']['xmin'] = int(float(value['bndbox']['xmin']))
        value['bndbox']['ymax'] = int(float(value['bndbox']['ymax']))
        value['bndbox']['ymin'] = int(float(value['bndbox']['ymin']))
    if not value:
        return key, []
    return key, value

xml_files = glob.glob(
    f"{annotations_in}/*.xml",
    recursive=True)
for xml_file in xml_files:
    # Open the file and read the contents
    with open(xml_file, 'r', encoding='utf-8') as file:
        src_xml = file.read()
    my_dict = xmltodict.parse(src_xml, postprocessor=postprocessor)
    dist_xml = dict2xml(my_dict)
    with open(xml_file, 'w', encoding='utf-8') as file:
        file.write(dist_xml)