Source code for mmaction.models.common.sub_batchnorm3d
# Copyright (c) OpenMMLab. All rights reserved.
from copy import deepcopy
import torch
import torch.nn as nn
from mmaction.registry import MODELS
[docs]@MODELS.register_module()
class SubBatchNorm3D(nn.Module):
"""Sub BatchNorm3d splits the batch dimension into N splits, and run BN on
each of them separately (so that the stats are computed on each subset of
examples (1/N of batch) independently). During evaluation, it aggregates
the stats from all splits into one BN.
Args:
num_features (int): Dimensions of BatchNorm.
"""
def __init__(self, num_features, **cfg):
super(SubBatchNorm3D, self).__init__()
self.num_features = num_features
self.cfg_ = deepcopy(cfg)
self.num_splits = self.cfg_.pop('num_splits', 1)
self.num_features_split = self.num_features * self.num_splits
# only keep one set of affine params, not in .bn or .split_bn
self.cfg_['affine'] = False
self.bn = nn.BatchNorm3d(num_features, **self.cfg_)
self.split_bn = nn.BatchNorm3d(self.num_features_split, **self.cfg_)
self.init_weights(cfg)
[docs] def init_weights(self, cfg):
"""Initialize weights."""
if cfg.get('affine', True):
self.weight = torch.nn.Parameter(torch.ones(self.num_features))
self.bias = torch.nn.Parameter(torch.zeros(self.num_features))
self.affine = True
else:
self.affine = False
def _get_aggregated_mean_std(self, means, stds, n):
"""Calculate aggregated mean and std."""
mean = means.view(n, -1).sum(0) / n
std = stds.view(n, -1).sum(0) / n + (
(means.view(n, -1) - mean)**2).view(n, -1).sum(0) / n
return mean.detach(), std.detach()
[docs] def aggregate_stats(self):
"""Synchronize running_mean, and running_var to self.bn.
Call this before eval, then call model.eval(); When eval, forward
function will call self.bn instead of self.split_bn, During this time
the running_mean, and running_var of self.bn has been obtained from
self.split_bn.
"""
if self.split_bn.track_running_stats:
aggre_func = self._get_aggregated_mean_std
self.bn.running_mean.data, self.bn.running_var.data = aggre_func(
self.split_bn.running_mean, self.split_bn.running_var,
self.num_splits)
self.bn.num_batches_tracked = self.split_bn.num_batches_tracked.detach(
)
[docs] def forward(self, x):
"""Defines the computation performed at every call."""
if self.training:
n, c, t, h, w = x.shape
assert n % self.num_splits == 0
x = x.view(n // self.num_splits, c * self.num_splits, t, h, w)
x = self.split_bn(x)
x = x.view(n, c, t, h, w)
else:
x = self.bn(x)
if self.affine:
x = x * self.weight.view(-1, 1, 1, 1)
x = x + self.bias.view(-1, 1, 1, 1)
return x