Keras 错误：无法将符号 Keras 输入/输出转换为 numpy 数组

Posted 2023-02-16

【中文标题】Keras 错误：无法将符号 Keras 输入/输出转换为 numpy 数组

【英文标题】：Keras error: Cannot convert a symbolic Keras input/output to a numpy array

【发布时间】：2021-12-17 06:40:44

【问题描述】：

我有来自 RaGan（相对论平均 Gan）的这段代码（部分代码）：

def get_ragan_network(generator,discriminator,optimizer):
    imgs_hr = Input(image_shape)
    generated_hr = Input(image_shape)

    Discriminator_real_out = discriminator(imgs_hr)
    Discriminator_fake_out = discriminator(generated_hr)


    Real_Fake_relativistic_average_out = tf.add(Discriminator_real_out,-(K.mean(Discriminator_fake_out, axis=0)))
    Fake_Real_relativistic_average_out = tf.add(Discriminator_fake_out,-(K.mean(Discriminator_real_out, axis=0)))

    epsilon=0.000001 
    def relativistic_discriminator_loss(y_true, y_pred):
        if isinstance(Real_Fake_relativistic_average_out, np.ndarray):
            return -(K.mean(K.log(K.sigmoid(Real_Fake_relativistic_average_out)+epsilon ),axis=0)
                 +K.mean(K.log(1-K.sigmoid(Fake_Real_relativistic_average_out)+epsilon),axis=0))
        else:
            return -(K.mean(K.log(K.sigmoid(Real_Fake_relativistic_average_out)+epsilon ),axis=0)
                 +K.mean(K.log(1-K.sigmoid(Fake_Real_relativistic_average_out)+epsilon),axis=0))

    model = Model([generated_hr,imgs_hr],[Discriminator_real_out,Discriminator_fake_out])

    model.compile(optimizer=optimizer, loss=[relativistic_discriminator_loss,None])
    return model

但是当我执行代码时，我得到了这个错误：

Cannot convert a symbolic Keras input/output to a numpy array. This error may indicate that you're trying to pass a symbolic value to a NumPy call, which is not supported. Or, you may be trying to pass Keras symbolic inputs/outputs to a TF API that does not register dispatching, preventing Keras from automatically converting the API call to a lambda layer in the Functional Model.

我不知道如何解决。

Numpy 版本 = 1.19

张量流版本 = 2.6

【参考方案1】：

我认为错误来自这一行：isinstance(Real_Fake_relativistic_average_out, np.ndarray)，因为Real_Fake_relativistic_average_out 来自鉴别器输出，而np.ndarray 是一个 numpy 对象。显然，条件 if else 语句是无用的。

【讨论】：

不不，这不是错误，如果我删除条件 if else 我得到相同的错误。我已经从 DRaGan、RaGan 等中探索了其他旧示例......并且都给了我同样的错误。我认为问题可能是numpy版本或tensorflow版本，但我需要找到一个没有降级版本的解决方案。