为对象检测任务创建tfrecord

时间:2018-06-03 09:55:09

标签: tensorflow tfrecord

我使用tensorflow对象检测api为微调任务创建数据集。

我的目录结构是:

火车/

- imgs /

---- img1.jpg

- ann /

---- img1.csv

其中csv(每个图像一个)为label, x, y, w, h

我用这个脚本来保存tfrecord:

    import tensorflow as tf
    from os import listdir
    import os
    from os.path import isfile, join
    import csv
    import json

    from object_detection.utils import dataset_util


    flags = tf.app.flags
    flags.DEFINE_string('output_path', '', 'Path to output TFRecord')
    FLAGS = flags.FLAGS

    LABEL_DICT = {}
    counter = 0

    def create_tf_example(example):
      # TODO(user): Populate the following variables from your example.
      height = 404 # Image height
      width = 720 # Image width
      filename = example['path'].encode('utf-8').strip() # Filename of the image. Empty if image is not from file

      with tf.gfile.GFile(example['path'], 'rb') as fid:
        encoded_image_data = fid.read()

      image_format = 'jpeg'.encode('utf-8').strip() # b'jpeg' or b'png'

      xmins = [] # List of normalized left x coordinates in bounding box (1 per box)
      xmaxs = [] # List of normalized right x coordinates in bounding box
                 # (1 per box)
      ymins = [] # List of normalized top y coordinates in bounding box (1 per box)
      ymaxs = [] # List of normalized bottom y coordinates in bounding box
                 # (1 per box)
      classes_text = [] # List of string class name of bounding box (1 per box)
      classes = [] # List of integer class id of bounding box (1 per box)

      for box in example['boxes']:
        #if box['occluded'] is False:
        #print("adding box")
        xmins.append(float(int(box['x']) / width))
        xmaxs.append(float(int(box['w']) + int(box['x']) / width))
        ymins.append(float(int(box['y']) / height))
        ymaxs.append(float(int(box['h']) + int(box['y']) / height))
        classes_text.append(box['label'].encode('utf-8'))
        classes.append(int(LABEL_DICT[box['label']]))


      tf_example = tf.train.Example(features=tf.train.Features(feature={
          'image/height': dataset_util.int64_feature(height),
          'image/width': dataset_util.int64_feature(width),
          'image/filename': dataset_util.bytes_feature(filename),
          'image/source_id': dataset_util.bytes_feature(filename),
          'image/encoded': dataset_util.bytes_feature(encoded_image_data),
          'image/format': dataset_util.bytes_feature(image_format),
          'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
          'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
          'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
          'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
          'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
          'image/object/class/label': dataset_util.int64_list_feature(classes),
      }))

      return tf_example

    def ex_info(img_path, ann_path):
      boxes = []
      head = ['label','x','y','w','h']
      with open(ann_path, 'r') as csvfile:
        annreader = csv.DictReader(csvfile, fieldnames=head)
        for box in annreader:
          boxes.append(box)
          LABEL_DICT[box['label']] = LABEL_DICT.get(box['label'], len(LABEL_DICT) + 1)

      ex = {
        "path" : img_path,
        "boxes" : boxes
      }

      return ex

    def main(_):
      writer = tf.python_io.TFRecordWriter(FLAGS.output_path)

      # TODO(user): Write code to read in your dataset to examples variable
      dataset_dir = "train"
      ann_dir = join(dataset_dir, "ann")
      imgs_dir = join(dataset_dir, "imgs")
      labelDest = "tfTrain/data/labels_map.pbtxt"

      imgs = [join(imgs_dir, f) for f in listdir(imgs_dir) if isfile(join(imgs_dir, f))]
      anns = [join(ann_dir, os.path.basename(im).replace("jpg","csv")) for im in imgs]

      for img,ann in zip(imgs,anns):
        example = ex_info(img,ann)
        #tf_example = create_tf_example(example)
        #writer.write(tf_example.SerializeToString())


      with open(labelDest, 'w', encoding='utf-8') as outL:
        for name,key in LABEL_DICT.items():
          outL.write("item { \n  id: " + str(key) + "\n  name: '" + name + "'\n}\n")


      writer.close()


    if __name__ == '__main__':
      tf.app.run()

但是当我运行火车脚本时我得到了这个错误

  

python train.py --logtostderr --train_dir =。/ models / train   --pipeline_config_path = faster_rcnn_resnet101_coc           o.config
          警告:tensorflow:来自models / research / object_detection / trainer.py:257:create_global_step   (来自tensorflow.contrib.framewo           rk.python.ops.variables)已弃用,将在以后的版本中删除。           更新说明:           请切换到tf.train.create_global_step           Traceback(最近一次调用最后一次):             文件" models / research / object_detection / utils / label_map_util.py",第135行,   在load_labelmap中               text_format.Merge(label_map_string,label_map)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第525行,在Merge               descriptor_pool = descriptor_pool)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第579行,在MergeLines中               return parser.MergeLines(lines,message)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第612行,在MergeLines中               self._ParseOrMerge(行,消息)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第627行,在_ParseOrMerge中               self._MergeField(tokenizer,message)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第787行,在_MergeField中               合并(标记器,消息,字段)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第815行,在_MergeMes​​sageField中               self._MergeField(tokenizer,sub_message)             文件" /home/user/anaconda3/envs/tf/lib/python3.6/site-packages/google/protobuf/text_format.py",   第695行,在_MergeField中               (message_descriptor.full_name,name))           google.protobuf.text_format.ParseError:23:20:消息类型" object_detection.protos.StringIntLabelMapItem"没有命名的字段   " S"

     

在处理上述异常期间,发生了另一个异常:

    Traceback (most recent call last):
      File "train.py", line 184, in <module>
        tf.app.run()
      File "/home/user/anaconda3/envs/tf/lib/python3.6/site-packages/tensorflow/python/platform/app.py",
     

第126行,在运行中               _sys.exit(主(argv的))             文件&#34; train.py&#34;,第180行,在main中               graph_hook_fn = graph_rewriter_fn)             文件&#34; models / research / object_detection / trainer.py&#34;,第264行,列车               train_config.prefetch_queue_capacity,data_augmentation_options)             文件&#34; models / research / object_detection / trainer.py&#34;,第59行,在create_input_queue中               tensor_dict = create_tensor_dict_fn()             在get_next中输入文件&#34; train.py&#34;,第121行               dataset_builder.build(配置))。get_next()             File&#34; models / research / object_detection / builders / dataset_builder.py&#34;,line   155,在构建中               label_map_proto_file = label_map_proto_file)             文件&#34; models / research / object_detection / data_decoders / tf_example_decoder.py&#34;,   第245行,在 init 中               use_display_name)             文件&#34; models / research / object_detection / utils / label_map_util.py&#34;,第152行,   在get_label_map_dict中               label_map = load_labelmap(label_map_path)             文件&#34; models / research / object_detection / utils / label_map_util.py&#34;,第137行,   在load_labelmap中               label_map.ParseFromString(label_map_string)           TypeError:需要类似字节的对象,而不是&#39; str&#39;

     

我不明白这个问题是什么。在tfrecord?在里面   labels.pbtxt?还是在配置文件中?

1 个答案:

答案 0 :(得分:1)

好的,我刚刚解决了调试张量流问题。虽然采用utf-8格式,但是我的标签很难被张量流读取,因为有些奇怪的字符如&amp; ùà。从csv中删除让火车开始