Flask部署Keras模型遇到的问题

AttributeError: ‘_thread._local’ object has no attribute ‘value’

• 定位到 tensorflow_backend.py 这个文件

• 在这个位置

  @functools.wraps(func)
  def symbolic_fn_wrapper(*args, **kwargs):
      if _SYMBOLIC_SCOPE.value:
          with get_graph().as_default():
              return func(*args, **kwargs)
      else:
          return func(*args, **kwargs)
  return symbolic_fn_wrapper

• 增加

  _SYMBOLIC_SCOPE.value = True

• 变成这样

  @functools.wraps(func)
  def symbolic_fn_wrapper(*args, **kwargs):
      _SYMBOLIC_SCOPE.value = True
      if _SYMBOLIC_SCOPE.value:
          with get_graph().as_default():
              return func(*args, **kwargs)
      else:
          return func(*args, **kwargs)
  return symbolic_fn_wrapper

• 问题解决