我正在尝试通过执行export_inference_graph.py脚本导出模型。
我尝试使用经过训练的 model.ckpt 和 ssd_mobilenet_v1_pets 的官方示例文件。
在 cmd 中输入:
python export_inference_graph.py \ --input_type image_tensor \ --pipeline_config_path training/ssd_mobilenet_v1_pets.config \ --trained_checkpoint_prefix training/model.ckpt-2453 \ --output_directory heart_graph
我使用的是 TensorFlow 1.4，但总是收到以下错误:
Traceback (most recent call last):
File "export_inference_graph.py", line 119, in <module>
tf.app.run()
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\platform\app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "export_inference_graph.py", line 115, in main
FLAGS.output_directory, input_shape)
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\Lib\site-packages\tensorflow\models\research\object_detection\exporter.py", line 427, in export_inference_graph
input_shape, optimize_graph, output_collection_name)
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\Lib\site-packages\tensorflow\models\research\object_detection\exporter.py", line 353, in _export_inference_graph
postprocessed_tensors = detection_model.postprocess(output_tensors)
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\Lib\site-packages\tensorflow\models\research\object_detection\meta_architectures\ssd_meta_arch.py", line 405, in postprocess
class_predictions_without_background)
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\Lib\site-packages\tensorflow\models\research\object_detection\builders\post_processing_builder.py", line 94, in score_converter_fn
scaled_logits = tf.divide(logits, logit_scale, name='scale_logits')
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\ops\math_ops.py", line 309, in divide
return DivideDelegateWithName(x, name) / y
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\ops\math_ops.py", line 294, in __truediv__
return _truediv_python3(self.x, y, self.name)
File "C:\Users\<Name>\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\ops\math_ops.py", line 981, in _truediv_python3
(x_dtype, y_dtype))
TypeError: x and y must have the same dtype, got tf.float32 != tf.int32
问题出在哪里以及如何解决？
最佳答案
最近我使用对象检测做了一些有趣的事情，也出现了这个错误，我在github上查找问题并找到解决方案。
在https://github.com/tensorflow/models/issues/2774中，有一个更改源代码的解决方案，我尝试一下。它有效!!
您可以找到 post_processing_builder.py 并使用以下命令更改函数
def _score_converter_fn_with_logit_scale(tf_score_converter_fn, logit_scale):
"""创建一个函数来缩放 logits，然后应用 Tensorflow 函数。"""
def Score_converter_fn(logits):
cr = logit_scale
cr = tf.constant([[cr]],tf.float32)
打印(logit_scale)
打印(逻辑)
scaled_logits = tf.divide(logits, cr, name='scale_logits') #更改 logit_scale
返回 tf_score_converter_fn(scaled_logits, name='convert_scores')
Score_converter_fn.__name__ = '%s_with_logit_scale' % (
tf_score_converter_fn.__name__)
返回score_converter_fn
然后进入research文件夹，运行
python setup.py 安装
那就OK了!
顺便问一下，我不知道你是否应该重新安装research文件夹中的slim，你最好也重新安装一下。
关于python - 运行export_inference_graph.py会给出类型差异错误，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/47242485/