ValueError：无法使用来自 keras 的损失函数将权重广播到值

Question

I'm using tensorflow keras to make a simple CNN_3D model.

inputs = keras.Input(shape=(65, 65, 65, 1), name='t1_image')
x = layers.Conv3D(16, (4, 4, 4), name='cnn_1')(inputs)
x = layers.Dropout(0.3)(x)
x = layers.BatchNormalization()(x)
x = layers.LeakyReLU()(x)
x = layers.Conv3D(24, (3, 3, 3), name='cnn_2')(x)
x = layers.Dropout(0.3)(x)
x = layers.BatchNormalization()(x)
x = layers.LeakyReLU()(x)
x = layers.MaxPooling3D((2, 2, 2), name='max_pool_1')(x)
x = layers.Conv3D(28, (3, 3, 3), name='cnn_3')(x)
x = layers.Dropout(0.3)(x)
x = layers.BatchNormalization()(x)
x = layers.LeakyReLU()(x)
x = layers.MaxPooling3D((2, 2, 2), name='max_pool_2')(x)
x = layers.Conv3D(34, (4, 4, 4), name='cnn_4')(x)
x = layers.Dropout(0.3)(x)
x = layers.BatchNormalization()(x)
x = layers.LeakyReLU()(x)
x = layers.Conv3D(2, (4, 4, 4), name='cnn_5')(x)
x = layers.Dropout(0.3)(x)
x = layers.BatchNormalization()(x)
x = layers.LeakyReLU()(x)
outputs = layers.Dense(1, activation='sigmoid', name='predictions')(x)

#print(outputs.shape)

model = keras.Model(inputs=inputs, outputs=outputs)
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=2e-5),
              loss=tf.keras.losses.KLDivergence(), metrics=['accuracy'])

So from the debug message printing, the outputs shape is (None, 8, 8, 8, 1) and my label shape is also (8, 8, 8, 1). So basically I want to calculate the KLDivergence between two cubes.

However, I'm getting this error message;

Traceback (most recent call last):
  File "new_seg.py", line 136, in <module>
    loss=tf.keras.losses.KLDivergence(), metrics=['accuracy'])
  File "/N/soft/rhel7/deeplearning/Python-3.7.6/lib/python3.7/site-packages/keras/backend/tensorflow_backend.py", line 75, in symbolic_fn_wrapper
    return func(*args, **kwargs)
  File "/N/soft/rhel7/deeplearning/Python-3.7.6/lib/python3.7/site-packages/keras/engine/training.py", line 229, in compile
    self.total_loss = self._prepare_total_loss(masks)
  File "/N/soft/rhel7/deeplearning/Python-3.7.6/lib/python3.7/site-packages/keras/engine/training.py", line 692, in _prepare_total_loss
    y_true, y_pred, sample_weight=sample_weight)
  File "/N/u/jp109/Carbonate/.local/lib/python3.7/site-packages/tensorflow_core/python/keras/losses.py", line 128, in __call__
    losses, sample_weight, reduction=self._get_reduction())
  File "/N/u/jp109/Carbonate/.local/lib/python3.7/site-packages/tensorflow_core/python/keras/utils/losses_utils.py", line 107, in compute_weighted_loss
    losses, sample_weight)
  File "/N/u/jp109/Carbonate/.local/lib/python3.7/site-packages/tensorflow_core/python/ops/losses/util.py", line 148, in scale_losses_by_sample_weight
    sample_weight = weights_broadcast_ops.broadcast_weights(sample_weight, losses)
  File "/N/u/jp109/Carbonate/.local/lib/python3.7/site-packages/tensorflow_core/python/ops/weights_broadcast_ops.py", line 167, in broadcast_weights
    with ops.control_dependencies((assert_broadcastable(weights, values),)):
  File "/N/u/jp109/Carbonate/.local/lib/python3.7/site-packages/tensorflow_core/python/ops/weights_broadcast_ops.py", line 103, in assert_broadcastable
    weights_rank_static, values.shape, weights.shape))
ValueError: weights can not be broadcast to values. values.rank=4. weights.rank=1. values.shape=(None, 8, 8, 8). weights.shape=(None,).

I'm guessing the important line is this;

ValueError: weights can not be broadcast to values. values.rank=4. weights.rank=1. values.shape=(None, 8, 8, 8). weights.shape=(None,).

which comes from this line;

model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=2e-5),
                  loss=tf.keras.losses.KLDivergence(), metrics=['accuracy'])

I don't understand what role weights is playing here and why the loss function is not working.

Does anybody know or have any suggestions about this issue?

Answer 1

You are mixing keras and tf.keras , you cannot do that.

Either you use only keras , or you use only tf.keras . Must choose one.