def sample_from_instances(inputs, outputs, num_samples):
"""Samples equally from instances."""
input_tensors = {}
output_tensors = {}
input_tensors[
standard_fields.InputDataFields.object_instance_id_image] = tf.reshape(
inputs[standard_fields.InputDataFields.object_instance_id_image],
[-1])
input_tensors[standard_fields.InputDataFields.
object_instance_id_image] = isu.map_labels_to_0_to_n(
input_tensors[standard_fields.InputDataFields.
object_instance_id_image])
seed_indices = isu.randomly_select_n_points_per_segment(
labels=input_tensors[
standard_fields.InputDataFields.object_instance_id_image],
num_points=num_samples,
include_ignore_label=False)
seed_indices = tf.reshape(seed_indices, [-1])
for field in standard_fields.get_input_image_fields():
if field in inputs:
input_tensors[field] = tf.gather(inputs[field], seed_indices)
for field in standard_fields.get_output_image_fields():
if field in outputs:
output_tensors[field] = tf.gather(outputs[field], seed_indices)
return input_tensors, output_tensors
def test_map_labels_to_0_to_n2(self):
labels = tf.constant([[-1, 1, 2],
[1, 1, 2]], dtype=tf.int32)
labels_0_n = isu.map_labels_to_0_to_n(labels)
expected_labels_0_n = tf.constant([[-1, 0, 1],
[0, 0, 1]], dtype=tf.int32)
self.assertAllEqual(labels_0_n.numpy(), expected_labels_0_n.numpy())
def sample_from_labels_balanced(labels, num_samples, epsilon=1e-5):
"""Samples from labels inverse proportional to their frequency.
Args:
labels: A tf.int32 tensor of size [N].
num_samples: Number of samples.
epsilon: A very small number.
Returns:
indices: A tf.int32 tensor of size [num_samples].
"""
labels = isu.map_labels_to_0_to_n(labels)
max_label = tf.reduce_max(labels)
labels_onehot = tf.one_hot(labels, depth=(max_label + 1), dtype=tf.float32)
frequencies = tf.reduce_sum(labels_onehot, axis=0)
inv_frequencies = 1.0 / (frequencies + epsilon)
inv_frequencies = tf.gather(inv_frequencies, labels)
return tf.squeeze(tf.random.categorical(
tf.math.log(tf.expand_dims(inv_frequencies, axis=0)), num_samples),
axis=0)
