objective.py

# coding=utf-8
# Copyright 2020 The SimCLR Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific simclr governing permissions and
# limitations under the License.
# ==============================================================================
"""Contrastive loss functions."""

from absl import flags

import tensorflow.compat.v2 as tf

FLAGS = flags.FLAGS

LARGE_NUM = 1e9


def add_supervised_loss(labels, logits):
  """Compute mean supervised loss over local batch."""
  losses = tf.keras.losses.CategoricalCrossentropy(
      from_logits=True, reduction=tf.keras.losses.Reduction.NONE)(labels,
                                                                  logits)
  return tf.reduce_mean(losses)


def add_contrastive_loss(hidden,
                         hidden_norm=True,
                         temperature=1.0,
                         strategy=None):
  """Compute loss for model.

  Args:
    hidden: hidden vector (`Tensor`) of shape (bsz, dim).
    hidden_norm: whether or not to use normalization on the hidden vector.
    temperature: a `floating` number for temperature scaling.
    strategy: context information for tpu.

  Returns:
    A loss scalar.
    The logits for contrastive prediction task.
    The labels for contrastive prediction task.
  """
  # Get (normalized) hidden1 and hidden2.
  if hidden_norm:
    hidden = tf.math.l2_normalize(hidden, -1)
  hidden1, hidden2 = tf.split(hidden, 2, 0)
  batch_size = tf.shape(hidden1)[0]

  # Gather hidden1/hidden2 across replicas and create local labels.
  if strategy is not None:
    hidden1_large = tpu_cross_replica_concat(hidden1, strategy)
    hidden2_large = tpu_cross_replica_concat(hidden2, strategy)
    enlarged_batch_size = tf.shape(hidden1_large)[0]
    # TODO(iamtingchen): more elegant way to convert u32 to s32 for replica_id.
    replica_context = tf.distribute.get_replica_context()
    replica_id = tf.cast(
        tf.cast(replica_context.replica_id_in_sync_group, tf.uint32), tf.int32)
    labels_idx = tf.range(batch_size) + replica_id * batch_size
    labels = tf.one_hot(labels_idx, enlarged_batch_size * 2)
    masks = tf.one_hot(labels_idx, enlarged_batch_size)
  else:
    hidden1_large = hidden1
    hidden2_large = hidden2
    labels = tf.one_hot(tf.range(batch_size), batch_size * 2)
    masks = tf.one_hot(tf.range(batch_size), batch_size)

  logits_aa = tf.matmul(hidden1, hidden1_large, transpose_b=True) / temperature
  logits_aa = logits_aa - masks * LARGE_NUM
  logits_bb = tf.matmul(hidden2, hidden2_large, transpose_b=True) / temperature
  logits_bb = logits_bb - masks * LARGE_NUM
  logits_ab = tf.matmul(hidden1, hidden2_large, transpose_b=True) / temperature
  logits_ba = tf.matmul(hidden2, hidden1_large, transpose_b=True) / temperature

  loss_a = tf.nn.softmax_cross_entropy_with_logits(
      labels, tf.concat([logits_ab, logits_aa], 1))
  loss_b = tf.nn.softmax_cross_entropy_with_logits(
      labels, tf.concat([logits_ba, logits_bb], 1))
  loss = tf.reduce_mean(loss_a + loss_b)

  return loss, logits_ab, labels


def tpu_cross_replica_concat(tensor, strategy=None):
  """Reduce a concatenation of the `tensor` across TPU cores.

  Args:
    tensor: tensor to concatenate.
    strategy: A `tf.distribute.Strategy`. If not set, CPU execution is assumed.

  Returns:
    Tensor of the same rank as `tensor` with first dimension `num_replicas`
    times larger.
  """
  if strategy is None or strategy.num_replicas_in_sync <= 1:
    return tensor

  num_replicas = strategy.num_replicas_in_sync

  replica_context = tf.distribute.get_replica_context()
  with tf.name_scope('tpu_cross_replica_concat'):
    # This creates a tensor that is like the input tensor but has an added
    # replica dimension as the outermost dimension. On each replica it will
    # contain the local values and zeros for all other values that need to be
    # fetched from other replicas.
    ext_tensor = tf.scatter_nd(
        indices=[[replica_context.replica_id_in_sync_group]],
        updates=[tensor],
        shape=tf.concat([[num_replicas], tf.shape(tensor)], axis=0))

    # As every value is only present on one replica and 0 in all others, adding
    # them all together will result in the full tensor on all replicas.
    ext_tensor = replica_context.all_reduce(tf.distribute.ReduceOp.SUM,
                                            ext_tensor)

    # Flatten the replica dimension.
    # The first dimension size will be: tensor.shape[0] * num_replicas
    # Using [-1] trick to support also scalar input.
    return tf.reshape(ext_tensor, [-1] + ext_tensor.shape.as_list()[2:])