Keras 中的多个输出给出值错误答案

【问题标题】：Multiple outputs in Keras gives value errorKeras 中的多个输出给出值错误
【发布时间】：2016-12-18 16:39:58
【问题描述】：

我正在对 U-net 进行语义分割修改。

我有两个来自网络的输出：

    model = Model(input=inputs, output= [conv10, dense3])
    model.compile(optimizer=Adam(lr=1e-5), loss=common_loss, metrics=[common_loss])

常见损失定义为：

def common_loss(y_true, y_pred):
    segmentation_loss  = categorical_crossentropy(y_true[0], y_pred[0])
    classifiction_loss = categorical_crossentropy(y_true[1], y_pred[1])
    return segmentation_loss + alpha * classifiction_loss

当我运行这个时，我得到一个值错误：

File "y-net.py", line 138, in <module>
    train_and_predict()
File "y-net.py", line 133, in train_and_predict
    callbacks=[model_checkpoint], validation_data=(X_val, [y_img_val, y_class_val]))
  File "/home/gpu_users/meetshah/miniconda2/envs/check/lib/python2.7/site-packages/keras/engine/training.py", line 1124, in fit
    callback_metrics=callback_metrics)
  File "/home/gpu_users/meetshah/miniconda2/envs/check/lib/python2.7/site-packages/keras/engine/training.py", line 848, in _fit_loop
    callbacks.on_batch_end(batch_index, batch_logs)
  File "/home/gpu_users/meetshah/miniconda2/envs/check/lib/python2.7/site-packages/keras/callbacks.py", line 63, in on_batch_end
    callback.on_batch_end(batch, logs)
  File "/home/gpu_users/meetshah/miniconda2/envs/check/lib/python2.7/site-packages/keras/callbacks.py", line 191, in on_batch_end
    self.progbar.update(self.seen, self.log_values)
  File "/home/gpu_users/meetshah/miniconda2/envs/check/lib/python2.7/site-packages/keras/utils/generic_utils.py", line 147, in update
    if abs(avg) > 1e-3:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

我的实现和整个跟踪可以在这里找到：

https://gist.github.com/meetshah1995/19d54270e8d1b20f814e6c1495facc6a

【问题讨论】：

我从model.compile 中删除了该指标，它起作用了。显然 keras 还不支持多个输入指标。

标签： tensorflow deep-learning keras conv-neural-network

【解决方案1】：

您可以在此处查看如何实现具有多个输出的多个指标：https://github.com/EdwardTyantov/ultrasound-nerve-segmentation/blob/master/u_model.py。

    model.compile(optimizer=optimizer,
              loss={'main_output': dice_coef_loss, 'aux_output': 'binary_crossentropy'},
              metrics={'main_output': dice_coef, 'aux_output': 'acc'},
              loss_weights={'main_output': 1., 'aux_output': 0.5})

我不确定是否还支持组合输出指标。

【讨论】：

多个指标在 keras github repo 上仍然是一个悬而未决的问题。