File size: 5,103 Bytes
2c26ac8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
#!/usr/bin/env python3
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.

"""ResNe(X)t Head helper."""

import torch.nn as nn


class X3DHead(nn.Module):
    """
    X3D head.
    This layer performs a fully-connected projection during training, when the
    input size is 1x1x1. It performs a convolutional projection during testing
    when the input size is larger than 1x1x1. If the inputs are from multiple
    different pathways, the inputs will be concatenated after pooling.
    """

    def __init__(
        self,
        dim_in,
        dim_inner,
        dim_out,
        num_classes,
        pool_size,
        dropout_rate=0.0,
        act_func="softmax",
        inplace_relu=True,
        eps=1e-5,
        bn_mmt=0.1,
        norm_module=nn.BatchNorm3d,
        bn_lin5_on=False,
    ):
        """
        The `__init__` method of any subclass should also contain these
            arguments.
        X3DHead takes a 5-dim feature tensor (BxCxTxHxW) as input.

        Args:
            dim_in (float): the channel dimension C of the input.
            num_classes (int): the channel dimensions of the output.
            pool_size (float): a single entry list of kernel size for
                spatiotemporal pooling for the TxHxW dimensions.
            dropout_rate (float): dropout rate. If equal to 0.0, perform no
                dropout.
            act_func (string): activation function to use. 'softmax': applies
                softmax on the output. 'sigmoid': applies sigmoid on the output.
            inplace_relu (bool): if True, calculate the relu on the original
                input without allocating new memory.
            eps (float): epsilon for batch norm.
            bn_mmt (float): momentum for batch norm. Noted that BN momentum in
                PyTorch = 1 - BN momentum in Caffe2.
            norm_module (nn.Module): nn.Module for the normalization layer. The
                default is nn.BatchNorm3d.
            bn_lin5_on (bool): if True, perform normalization on the features
                before the classifier.
        """
        super(X3DHead, self).__init__()
        self.pool_size = pool_size
        self.dropout_rate = dropout_rate
        self.num_classes = num_classes
        self.act_func = act_func
        self.eps = eps
        self.bn_mmt = bn_mmt
        self.inplace_relu = inplace_relu
        self.bn_lin5_on = bn_lin5_on
        self._construct_head(dim_in, dim_inner, dim_out, norm_module)

    def _construct_head(self, dim_in, dim_inner, dim_out, norm_module):

        self.conv_5 = nn.Conv3d(
            dim_in,
            dim_inner,
            kernel_size=(1, 1, 1),
            stride=(1, 1, 1),
            padding=(0, 0, 0),
            bias=False,
        )
        self.conv_5_bn = norm_module(
            num_features=dim_inner, eps=self.eps, momentum=self.bn_mmt
        )
        self.conv_5_relu = nn.ReLU(self.inplace_relu)

        if self.pool_size is None:
            self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
        else:
            self.avg_pool = nn.AvgPool3d(self.pool_size, stride=1)

        self.lin_5 = nn.Conv3d(
            dim_inner,
            dim_out,
            kernel_size=(1, 1, 1),
            stride=(1, 1, 1),
            padding=(0, 0, 0),
            bias=False,
        )
        if self.bn_lin5_on:
            self.lin_5_bn = norm_module(
                num_features=dim_out, eps=self.eps, momentum=self.bn_mmt
            )
        self.lin_5_relu = nn.ReLU(self.inplace_relu)

        if self.dropout_rate > 0.0:
            self.dropout = nn.Dropout(self.dropout_rate)
        # Perform FC in a fully convolutional manner. The FC layer will be
        # initialized with a different std comparing to convolutional layers.
        self.projection = nn.Linear(dim_out, self.num_classes, bias=True)

        # Softmax for evaluation and testing.
        if self.act_func == "softmax":
            self.act = nn.Softmax(dim=4)
        elif self.act_func == "sigmoid":
            self.act = nn.Sigmoid()
        else:
            raise NotImplementedError(
                "{} is not supported as an activation" "function.".format(
                    self.act_func)
            )

    def forward(self, inputs):
        # In its current design the X3D head is only useable for a single
        # pathway input.
        assert len(inputs) == 1, "Input tensor does not contain 1 pathway"
        x = self.conv_5(inputs[0])
        x = self.conv_5_bn(x)
        x = self.conv_5_relu(x)
        x = self.avg_pool(x)

        x = self.lin_5(x)
        if self.bn_lin5_on:
            x = self.lin_5_bn(x)
        x = self.lin_5_relu(x)

        # (N, C, T, H, W) -> (N, T, H, W, C).
        x = x.permute((0, 2, 3, 4, 1))
        # Perform dropout.
        if hasattr(self, "dropout"):
            x = self.dropout(x)
        x = self.projection(x)

        # Performs fully convlutional inference.
        if not self.training:
            x = self.act(x)
            x = x.mean([1, 2, 3])

        x = x.view(x.shape[0], -1)
        return x