def do_3d_pooling(feature_matrix,
stride_length_px=2,
pooling_type_string=MAX_POOLING_TYPE_STRING):
"""Pools 3-D feature maps.
:param feature_matrix: Input feature maps (numpy array). Dimensions must be
M x N x H x C or 1 x M x N x H x C.
:param stride_length_px: See doc for `do_2d_pooling`.
:param pooling_type_string: Pooling type (must be accepted by
`_check_pooling_type`).
:return: feature_matrix: Output feature maps (numpy array). Dimensions will
be 1 x m x n x h x C.
"""
error_checking.assert_is_numpy_array_without_nan(feature_matrix)
error_checking.assert_is_integer(stride_length_px)
error_checking.assert_is_geq(stride_length_px, 2)
_check_pooling_type(pooling_type_string)
if len(feature_matrix.shape) == 4:
feature_matrix = numpy.expand_dims(feature_matrix, axis=0)
error_checking.assert_is_numpy_array(feature_matrix, num_dimensions=5)
feature_tensor = K.pool3d(x=K.variable(feature_matrix),
pool_mode=pooling_type_string,
pool_size=(stride_length_px, stride_length_px,
stride_length_px),
strides=(stride_length_px, stride_length_px,
stride_length_px),
padding='valid',
data_format='channels_last')
return feature_tensor.eval(session=K.get_session())
def loss_fn(y_true, y_pred):
y_pred = K.clip(y_pred, epsilon, 1 - epsilon)
cross_entropy = K.stack(
[-(y_true * K.log(y_pred)), -((1 - y_true) * K.log(1 - y_pred))],
axis=-1)
loss = cross_entropy * w
if boundary_weight is not None:
y_true_avg = K.pool3d(y_true,
pool_size=(pool, ) * 3,
padding='same',
pool_mode='avg')
boundaries = K.cast(y_true_avg >= epsilon, 'float32') \
* K.cast(y_true_avg <= 1 - epsilon, 'float32')
loss += cross_entropy * K.stack([boundaries, boundaries],
axis=-1) * boundary_weight
return K.mean(K.sum(loss, axis=-1))