def add_triplet_loss():
"""Adds triplet loss."""
anchor_keypoints_3d, positive_keypoints_3d = tf.unstack(
inputs[common_module.KEY_KEYPOINTS_3D], num=2, axis=1)
anchor_keypoint_masks_3d, positive_keypoint_masks_3d = None, None
if FLAGS.use_inferred_keypoint_masks_for_triplet_label:
anchor_keypoint_masks_2d, positive_keypoint_masks_2d = tf.unstack(
inputs[
common_module.KEY_PREPROCESSED_KEYPOINT_MASKS_2D],
num=2,
axis=1)
anchor_keypoint_masks_3d = keypoint_utils.transfer_keypoint_masks(
anchor_keypoint_masks_2d,
input_keypoint_profile=configs['keypoint_profile_2d'],
output_keypoint_profile=configs['keypoint_profile_3d'],
enforce_surjectivity=True)
positive_keypoint_masks_3d = keypoint_utils.transfer_keypoint_masks(
positive_keypoint_masks_2d,
input_keypoint_profile=configs['keypoint_profile_2d'],
output_keypoint_profile=configs['keypoint_profile_3d'],
enforce_surjectivity=True)
triplet_anchor_embeddings, triplet_positive_embeddings = tf.unstack(
pipeline_utils.stack_embeddings(
outputs, configs['triplet_embedding_keys']),
axis=1)
if FLAGS.use_normalized_embeddings_for_triplet_loss:
triplet_anchor_embeddings = tf.math.l2_normalize(
triplet_anchor_embeddings, axis=-1)
triplet_positive_embeddings = tf.math.l2_normalize(
triplet_positive_embeddings, axis=-1)
triplet_anchor_mining_embeddings, triplet_positive_mining_embeddings = (
tf.unstack(pipeline_utils.stack_embeddings(
outputs, configs['triplet_mining_embedding_keys']),
axis=1))
if FLAGS.use_normalized_embeddings_for_triplet_mining:
triplet_anchor_mining_embeddings = tf.math.l2_normalize(
triplet_anchor_mining_embeddings, axis=-1)
triplet_positive_mining_embeddings = tf.math.l2_normalize(
triplet_positive_mining_embeddings, axis=-1)
triplet_loss, triplet_loss_summaries = (
loss_utils.compute_keypoint_triplet_losses(
anchor_embeddings=triplet_anchor_embeddings,
positive_embeddings=triplet_positive_embeddings,
match_embeddings=triplet_positive_embeddings,
anchor_keypoints=anchor_keypoints_3d,
match_keypoints=positive_keypoints_3d,
margin=FLAGS.triplet_loss_margin,
min_negative_keypoint_distance=(
configs['min_negative_keypoint_distance']),
use_semi_hard=FLAGS.use_semi_hard_triplet_negatives,
exclude_inactive_triplet_loss=(
FLAGS.exclude_inactive_triplet_loss),
anchor_keypoint_masks=anchor_keypoint_masks_3d,
match_keypoint_masks=positive_keypoint_masks_3d,
embedding_sample_distance_fn=(
configs['triplet_embedding_sample_distance_fn']),
keypoint_distance_fn=configs['keypoint_distance_fn'],
anchor_mining_embeddings=
triplet_anchor_mining_embeddings,
positive_mining_embeddings=
triplet_positive_mining_embeddings,
match_mining_embeddings=
triplet_positive_mining_embeddings,
summarize_percentiles=FLAGS.summarize_percentiles))
tf.losses.add_loss(triplet_loss,
loss_collection=tf.GraphKeys.LOSSES)
summaries.update(triplet_loss_summaries)
summaries['train/triplet_loss'] = triplet_loss
def test_transfer_keypoint_masks_case_2(self):
# Shape = [2, 16].
input_keypoint_masks = tf.constant([
[
1.0, # NOSE
1.0, # NECK
1.0, # LEFT_SHOULDER
1.0, # RIGHT_SHOULDER
0.0, # LEFT_ELBOW
1.0, # RIGHT_ELBOW
1.0, # LEFT_WRIST
0.0, # RIGHT_WRIST
0.0, # SPINE
0.0, # PELVIS
1.0, # LEFT_HIP
0.0, # RIGHT_HIP
1.0, # LEFT_KNEE
1.0, # RIGHT_KNEE
0.0, # LEFT_ANKLE
0.0, # RIGHT_ANKLE
],
[
0.0, # NOSE
0.0, # NECK
0.0, # LEFT_SHOULDER
0.0, # RIGHT_SHOULDER
1.0, # LEFT_ELBOW
0.0, # RIGHT_ELBOW
0.0, # LEFT_WRIST
1.0, # RIGHT_WRIST
1.0, # SPINE
1.0, # PELVIS
0.0, # LEFT_HIP
1.0, # RIGHT_HIP
0.0, # LEFT_KNEE
0.0, # RIGHT_KNEE
1.0, # LEFT_ANKLE
1.0, # RIGHT_ANKLE
]
])
input_keypoint_profile = keypoint_profiles.create_keypoint_profile_or_die(
'3DSTD16')
output_keypoint_profile = keypoint_profiles.create_keypoint_profile_or_die(
'2DSTD13')
# Shape = [2, 13].
output_keypoint_masks = keypoint_utils.transfer_keypoint_masks(
input_keypoint_masks, input_keypoint_profile, output_keypoint_profile)
self.assertAllClose(
output_keypoint_masks,
[
[
1.0, # NOSE_TIP
1.0, # LEFT_SHOULDER
1.0, # RIGHT_SHOULDER
0.0, # LEFT_ELBOW
1.0, # RIGHT_ELBOW
1.0, # LEFT_WRIST
0.0, # RIGHT_WRIST
1.0, # LEFT_HIP
0.0, # RIGHT_HIP
1.0, # LEFT_KNEE
1.0, # RIGHT_KNEE
0.0, # LEFT_ANKLE
0.0, # RIGHT_ANKLE
],
[
0.0, # NOSE_TIP
0.0, # LEFT_SHOULDER
0.0, # RIGHT_SHOULDER
1.0, # LEFT_ELBOW
0.0, # RIGHT_ELBOW
0.0, # LEFT_WRIST
1.0, # RIGHT_WRIST
0.0, # LEFT_HIP
1.0, # RIGHT_HIP
0.0, # LEFT_KNEE
0.0, # RIGHT_KNEE
1.0, # LEFT_ANKLE
1.0, # RIGHT_ANKLE
]
])
