Recieve list of all outputs as input to a custom loss function.

Important note: the development of Keras is currently taking place at github.com/tensorflow/tensorflow. Please file your bug report by creating a new issue in the TensorFlow repository.

Please note that multi-backend Keras development has been discontinued. Do not report issues about multi-backend Keras (Keras 2.3.1 and lower), only report issues about the TensorFlow implementation of Keras (tf.keras).

[manaspalaparthi]

I have the same problem too. my model outputs 3 tensors. model= Model(inputs = inputs,outputs =[predicted_image,depth,x_train])

my loss function should be

` def depth_loss_function(y_true, y_pred):

print(y_pred)

predict_image, depth , input_image = y_pred

# Point-wise depth
depth_loss = K.mean(K.abs(depth - y_true), axis=-1)

# image reconstruction
mse = tf.keras.losses.MeanSquaredError()
image_loss = mse(input_image - predict_image)

# Weights
w1 = 1.0
w2 = 1.0

return  (w2 * depth_loss) + (w2 * image_loss )`

compile method

model.compile(optimizer= optimizer, loss =depth_loss_function)

error 1 : y_pred is only first tensor. error 2 : by using predict_image, depth , input_image = y_pred Tensor objects are only iterable when eager execution is enabled. To iterate over this tensor use tf.map_fn.

[vladhondru25]

Yes, I created a custom keras layer (as last layer in my function), and inside it I used self.add_loss function (you can pass any function with any number of parameters in add_loss).