def testReturnsCorrectAnchorWiseLossWithLogitScaleSetting(self):
logit_scale = 100.
prediction_tensor = tf.constant(
[[[-100, 100, -100], [100, -100, -100], [0, 0, -100],
[-100, -100, 100]],
[[-100, 0, 0], [-100, 100, -100], [-100, 100, -100],
[100, -100, -100]]], tf.float32)
target_tensor = tf.constant([[[-100, 100, -100], [100, -100, -100],
[0, 0, -100], [-100, -100, 100]],
[[-100, 0, 0], [-100, 100, -100],
[-100, 100, -100], [100, -100, -100]]],
tf.float32)
weights = tf.constant([[1, 1, .5, 1], [1, 1, 1, 0]], tf.float32)
loss_op = losses.WeightedSoftmaxClassificationAgainstLogitsLoss(
logit_scale=logit_scale)
loss = loss_op(prediction_tensor, target_tensor, weights=weights)
# find softmax of the two prediction types above
softmax_pred1 = [np.exp(-1), np.exp(-1), np.exp(1)]
softmax_pred1 /= sum(softmax_pred1)
softmax_pred2 = [np.exp(0), np.exp(0), np.exp(-1)]
softmax_pred2 /= sum(softmax_pred2)
# compute the expected cross entropy for perfect matches
exp_entropy1 = sum([-x * np.log(x) for x in softmax_pred1])
exp_entropy2 = sum([-x * np.log(x) for x in softmax_pred2])
# weighted expected losses
exp_loss = np.matrix(
[[exp_entropy1, exp_entropy1, exp_entropy2 * .5, exp_entropy1],
[exp_entropy2, exp_entropy1, exp_entropy1, 0.]])
with self.test_session() as sess:
loss_output = sess.run(loss)
self.assertAllClose(loss_output, exp_loss)
def build_faster_rcnn_classification_loss(loss_config):
"""Builds a classification loss for Faster RCNN based on the loss config.
Args:
loss_config: A losses_pb2.ClassificationLoss object.
Returns:
Loss based on the config.
Raises:
ValueError: On invalid loss_config.
"""
if not isinstance(loss_config, losses_pb2.ClassificationLoss):
raise ValueError(
'loss_config not of type losses_pb2.ClassificationLoss.')
loss_type = loss_config.WhichOneof('classification_loss')
if loss_type == 'weighted_sigmoid':
return losses.WeightedSigmoidClassificationLoss()
if loss_type == 'weighted_softmax':
config = loss_config.weighted_softmax
return losses.WeightedSoftmaxClassificationLoss(
logit_scale=config.logit_scale)
if loss_type == 'weighted_logits_softmax':
config = loss_config.weighted_logits_softmax
return losses.WeightedSoftmaxClassificationAgainstLogitsLoss(
logit_scale=config.logit_scale)
# By default, Faster RCNN second stage classifier uses Softmax loss
# with anchor-wise outputs.
config = loss_config.weighted_softmax
return losses.WeightedSoftmaxClassificationLoss(
logit_scale=config.logit_scale)
def _build_classification_loss(loss_config):
"""Builds a classification loss based on the loss config.
Args:
loss_config: A losses_pb2.ClassificationLoss object.
Returns:
Loss based on the config.
Raises:
ValueError: On invalid loss_config.
"""
if not isinstance(loss_config, losses_pb2.ClassificationLoss):
raise ValueError(
'loss_config not of type losses_pb2.ClassificationLoss.')
loss_type = loss_config.WhichOneof('classification_loss')
if loss_type == 'weighted_sigmoid':
return losses.WeightedSigmoidClassificationLoss()
if loss_type == 'weighted_sigmoid_focal':
config = loss_config.weighted_sigmoid_focal
alpha = None
if config.HasField('alpha'):
alpha = config.alpha
return losses.SigmoidFocalClassificationLoss(gamma=config.gamma,
alpha=alpha)
if loss_type == 'weighted_softmax':
config = loss_config.weighted_softmax
return losses.WeightedSoftmaxClassificationLoss(
logit_scale=config.logit_scale)
if loss_type == 'weighted_logits_softmax':
config = loss_config.weighted_logits_softmax
return losses.WeightedSoftmaxClassificationAgainstLogitsLoss(
logit_scale=config.logit_scale)
if loss_type == 'bootstrapped_sigmoid':
config = loss_config.bootstrapped_sigmoid
return losses.BootstrappedSigmoidClassificationLoss(
alpha=config.alpha,
bootstrap_type=('hard' if config.hard_bootstrap else 'soft'))
raise ValueError('Empty loss config.')
def testReturnsCorrectAnchorWiseLoss(self):
prediction_tensor = tf.constant(
[[[-100, 100, -100], [100, -100, -100], [0, 0, -100],
[-100, -100, 100]],
[[-100, 0, 0], [-100, 100, -100], [-100, 100, -100],
[100, -100, -100]]], tf.float32)
target_tensor = tf.constant([[[-100, 100, -100], [100, -100, -100],
[100, -100, -100], [-100, -100, 100]],
[[-100, -100, 100], [-100, 100, -100],
[-100, 100, -100], [100, -100, -100]]],
tf.float32)
weights = tf.constant([[1, 1, .5, 1], [1, 1, 1, 0]], tf.float32)
loss_op = losses.WeightedSoftmaxClassificationAgainstLogitsLoss()
loss = loss_op(prediction_tensor, target_tensor, weights=weights)
exp_loss = np.matrix([[0, 0, -0.5 * math.log(.5), 0],
[-math.log(.5), 0, 0, 0]])
with self.test_session() as sess:
loss_output = sess.run(loss)
self.assertAllClose(loss_output, exp_loss)