Question

我尝试使用TensorFlow对象检测API重新训练模型。在文件pipeline.config中，我进行了更改：

路径
num_classes：1
batch_size：1（用于训练速度）
num_steps：1000（用于训练速度）

我在做

C:\Distr\models-master\models-master\research\object_detection>python legacy/train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/pipeline.config

训练成功，但是训练后pipeline.config为空。我有一个警告：

I0913 12:14:47.846972 28680 learning.py:507] global step 995: loss = 6.8707 (0.798 sec/step)
INFO:tensorflow:global step 996: loss = 2.7868 (0.820 sec/step)
I0913 12:14:48.669283 28680 learning.py:507] global step 996: loss = 2.7868 (0.820 sec/step)
INFO:tensorflow:global step 997: loss = 3.4453 (0.797 sec/step)
I0913 12:14:49.466856 28680 learning.py:507] global step 997: loss = 3.4453 (0.797 sec/step)
INFO:tensorflow:global step 998: loss = 5.1915 (0.804 sec/step)
I0913 12:14:50.271611 28680 learning.py:507] global step 998: loss = 5.1915 (0.804 sec/step)
INFO:tensorflow:global step 999: loss = 7.8720 (0.950 sec/step)
I0913 12:14:51.224293 28680 learning.py:507] global step 999: loss = 7.8720 (0.950 sec/step)
INFO:tensorflow:Recording summary at step 999.
I0913 12:14:51.677052 21604 supervisor.py:1050] Recording summary at step 999.
INFO:tensorflow:global_step/sec: 1.20015
I0913 12:14:52.015100 25292 supervisor.py:1099] global_step/sec: 1.20015
INFO:tensorflow:global step 1000: loss = 3.9892 (0.837 sec/step)
I0913 12:14:52.063099 28680 learning.py:507] global step 1000: loss = 3.9892 (0.837 sec/step)
INFO:tensorflow:Stopping Training.
I0913 12:14:52.064100 28680 learning.py:777] Stopping Training.
INFO:tensorflow:Finished training! Saving model to disk.
I0913 12:14:52.064100 28680 learning.py:785] Finished training! Saving model to disk.
C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\summary\writer\writer.py:386: UserWarning: Attempting to use a closed FileWriter. The operation will be a noop unless the FileWriter is explicitly reopened.
  warnings.warn("Attempting to use a closed FileWriter. "
enter code here

当我尝试导出模型时出现错误。我使用命令：

python export_inference_graph.py --input_type image_tensor --pipeline_config_path training/pipeline.config --trained_checkpoint_prefix training/model.ckpt-1000 --output_directory inference_graph

错误：

Traceback (most recent call last):
  File "export_inference_graph.py", line 162, in <module>
    tf.app.run()
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\platform\app.py", line 40, in run
    _run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef)
  File "C:\ProgramData\Anaconda3\lib\site-packages\absl\app.py", line 299, in run
    _run_main(main, args)
  File "C:\ProgramData\Anaconda3\lib\site-packages\absl\app.py", line 250, in _run_main
    sys.exit(main(argv))
  File "export_inference_graph.py", line 158, in main
    write_inference_graph=FLAGS.write_inference_graph)
  File "C:\Distr\models-master\models-master\research\object_detection\exporter.py", line 473, in export_inference_graph
    is_training=False)
  File "C:\Distr\models-master\models-master\research\object_detection\builders\model_builder.py", line 136, in build
    raise ValueError('Unknown meta architecture: {}'.format(meta_architecture))
ValueError: Unknown meta architecture: None

如果我最初更改空的pipeline.config，则导出工作成功。

在培训后要获取正确的pipeline.config文件需要解决哪些问题？

TensorFlow对象检测API训练模型后，文件pipeline.config为空

0 个答案: