def test_tf_data_feature_label_keys(self):
"""Tests the ability of a get_tf_data_datasets to have extra labels/key.
Test is done here because TAP is off in specific dataset tests.
"""
features_data = data_provider.get_tf_data_dataset(
dataset_name='scannet_scene',
split_name='val',
batch_size=1,
preprocess_fn=None,
is_training=True,
num_readers=2,
num_parallel_batches=2,
shuffle_buffer_size=2)
features = next(iter(features_data))
self.assertEqual(
features['mesh/vertices/positions'].get_shape().as_list()[2], 3)
self.assertEqual(
features['mesh/vertices/normals'].get_shape().as_list()[2], 3)
self.assertEqual(
features['mesh/vertices/colors'].get_shape().as_list()[2], 4)
self.assertEqual(
features['mesh/faces/polygons'].get_shape().as_list()[2], 3)
self.assertEqual(
features['mesh/vertices/semantic_labels'].get_shape().as_list()[2],
1)
self.assertEqual(
features['mesh/vertices/instance_labels'].get_shape().as_list()[2],
1)
def test_get_tf_data_dataset_tfrecord(self):
dataset = data_provider.get_tf_data_dataset(
dataset_name='waymo_object_per_frame',
split_name='val',
batch_size=1,
is_training=True,
preprocess_fn=None,
feature_keys=None,
label_keys=None,
num_readers=1,
filenames_shuffle_buffer_size=2,
num_epochs=0,
read_block_length=1,
shuffle_buffer_size=2,
num_parallel_batches=1,
num_prefetch_batches=1,
dataset_format='tfrecord',
)
tfrecord_features = next(iter(dataset))
self.assertAllEqual(
tfrecord_features['cameras/front/extrinsics/R'].shape, [1, 3, 3])
def test_tf_data_feature_label_keys(self):
"""Tests the ability of a get_tf_data_datasets to have extra labels/key.
Test is done here because TAP is off in specific dataset tests.
"""
features_data = data_provider.get_tf_data_dataset(
dataset_name='waymo_object_per_frame',
split_name='val',
batch_size=1,
preprocess_fn=None,
is_training=True,
num_readers=2,
num_parallel_batches=2,
shuffle_buffer_size=2)
features = next(iter(features_data))
cameras = ['front', 'front_left', 'front_right', 'side_left', 'side_right']
lidars = ['top', 'front', 'side_left', 'side_right', 'rear']
for camera in cameras:
self.assertAllEqual(
features[('cameras/%s/extrinsics/t' % camera)].get_shape().as_list(),
np.array([1, 3]))
self.assertAllEqual(
features[('cameras/%s/extrinsics/R' % camera)].get_shape().as_list(),
np.array([1, 3, 3]))
self.assertAllEqual(
features[('cameras/%s/intrinsics/distortion' %
camera)].get_shape().as_list(), np.array([1, 5]))
self.assertAllEqual(
features[('cameras/%s/intrinsics/K' % camera)].get_shape().as_list(),
np.array([1, 3, 3]))
self.assertAllEqual(
features[('cameras/%s/image' % camera)].get_shape().as_list()[3], 3)
for lidar in lidars:
self.assertEqual(
features[('lidars/%s/pointcloud/positions' %
lidar)].get_shape().as_list()[2], 3)
self.assertEqual(
features[('lidars/%s/pointcloud/intensity' %
lidar)].get_shape().as_list()[2], 1)
self.assertEqual(
features[('lidars/%s/pointcloud/elongation' %
lidar)].get_shape().as_list()[2], 1)
self.assertAllEqual(
features[('lidars/%s/extrinsics/R' % lidar)].get_shape().as_list(),
np.array([1, 3, 3]))
self.assertAllEqual(
features[('lidars/%s/extrinsics/t' % lidar)].get_shape().as_list(),
np.array([1, 3]))
self.assertEqual(
features['lidars/%s/camera_projections/positions' %
lidar].get_shape().as_list()[2], 2)
self.assertEqual(
features['lidars/%s/camera_projections/ids' %
lidar].get_shape().as_list()[2], 1)
self.assertEqual(features['objects/pose/R'].get_shape().as_list()[2], 3)
self.assertEqual(features['objects/pose/R'].get_shape().as_list()[3], 3)
self.assertEqual(features['objects/pose/t'].get_shape().as_list()[2], 3)
self.assertEqual(
features['objects/shape/dimension'].get_shape().as_list()[2], 3)
self.assertLen(features['objects/category/label'].get_shape().as_list(), 2)