File size: 9,137 Bytes
d7a991a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
from mmcv.cnn import (build_conv_layer, build_upsample_layer, constant_init,
                      normal_init)

from mmpose.models.builder import build_loss
from ..backbones.resnet import BasicBlock
from ..builder import HEADS


@HEADS.register_module()
class AEHigherResolutionHead(nn.Module):
    """Associative embedding with higher resolution head. paper ref: Bowen
    Cheng et al. "HigherHRNet: Scale-Aware Representation Learning for Bottom-
    Up Human Pose Estimation".

    Args:
        in_channels (int): Number of input channels.
        num_joints (int): Number of joints
        tag_per_joint (bool): If tag_per_joint is True,
            the dimension of tags equals to num_joints,
            else the dimension of tags is 1. Default: True
        extra (dict): Configs for extra conv layers. Default: None
        num_deconv_layers (int): Number of deconv layers.
            num_deconv_layers should >= 0. Note that 0 means
            no deconv layers.
        num_deconv_filters (list|tuple): Number of filters.
            If num_deconv_layers > 0, the length of
        num_deconv_kernels (list|tuple): Kernel sizes.
        cat_output (list[bool]): Option to concat outputs.
        with_ae_loss (list[bool]): Option to use ae loss.
        loss_keypoint (dict): Config for loss. Default: None.
    """

    def __init__(self,
                 in_channels,
                 num_joints,
                 tag_per_joint=True,
                 extra=None,
                 num_deconv_layers=1,
                 num_deconv_filters=(32, ),
                 num_deconv_kernels=(4, ),
                 num_basic_blocks=4,
                 cat_output=None,
                 with_ae_loss=None,
                 loss_keypoint=None):
        super().__init__()

        self.loss = build_loss(loss_keypoint)
        dim_tag = num_joints if tag_per_joint else 1

        self.num_deconvs = num_deconv_layers
        self.cat_output = cat_output

        final_layer_output_channels = []

        if with_ae_loss[0]:
            out_channels = num_joints + dim_tag
        else:
            out_channels = num_joints

        final_layer_output_channels.append(out_channels)
        for i in range(num_deconv_layers):
            if with_ae_loss[i + 1]:
                out_channels = num_joints + dim_tag
            else:
                out_channels = num_joints
            final_layer_output_channels.append(out_channels)

        deconv_layer_output_channels = []
        for i in range(num_deconv_layers):
            if with_ae_loss[i]:
                out_channels = num_joints + dim_tag
            else:
                out_channels = num_joints
            deconv_layer_output_channels.append(out_channels)

        self.final_layers = self._make_final_layers(
            in_channels, final_layer_output_channels, extra, num_deconv_layers,
            num_deconv_filters)
        self.deconv_layers = self._make_deconv_layers(
            in_channels, deconv_layer_output_channels, num_deconv_layers,
            num_deconv_filters, num_deconv_kernels, num_basic_blocks,
            cat_output)

    @staticmethod
    def _make_final_layers(in_channels, final_layer_output_channels, extra,
                           num_deconv_layers, num_deconv_filters):
        """Make final layers."""
        if extra is not None and 'final_conv_kernel' in extra:
            assert extra['final_conv_kernel'] in [1, 3]
            if extra['final_conv_kernel'] == 3:
                padding = 1
            else:
                padding = 0
            kernel_size = extra['final_conv_kernel']
        else:
            kernel_size = 1
            padding = 0

        final_layers = []
        final_layers.append(
            build_conv_layer(
                cfg=dict(type='Conv2d'),
                in_channels=in_channels,
                out_channels=final_layer_output_channels[0],
                kernel_size=kernel_size,
                stride=1,
                padding=padding))

        for i in range(num_deconv_layers):
            in_channels = num_deconv_filters[i]
            final_layers.append(
                build_conv_layer(
                    cfg=dict(type='Conv2d'),
                    in_channels=in_channels,
                    out_channels=final_layer_output_channels[i + 1],
                    kernel_size=kernel_size,
                    stride=1,
                    padding=padding))

        return nn.ModuleList(final_layers)

    def _make_deconv_layers(self, in_channels, deconv_layer_output_channels,
                            num_deconv_layers, num_deconv_filters,
                            num_deconv_kernels, num_basic_blocks, cat_output):
        """Make deconv layers."""
        deconv_layers = []
        for i in range(num_deconv_layers):
            if cat_output[i]:
                in_channels += deconv_layer_output_channels[i]

            planes = num_deconv_filters[i]
            deconv_kernel, padding, output_padding = \
                self._get_deconv_cfg(num_deconv_kernels[i])

            layers = []
            layers.append(
                nn.Sequential(
                    build_upsample_layer(
                        dict(type='deconv'),
                        in_channels=in_channels,
                        out_channels=planes,
                        kernel_size=deconv_kernel,
                        stride=2,
                        padding=padding,
                        output_padding=output_padding,
                        bias=False), nn.BatchNorm2d(planes, momentum=0.1),
                    nn.ReLU(inplace=True)))
            for _ in range(num_basic_blocks):
                layers.append(nn.Sequential(BasicBlock(planes, planes), ))
            deconv_layers.append(nn.Sequential(*layers))
            in_channels = planes

        return nn.ModuleList(deconv_layers)

    @staticmethod
    def _get_deconv_cfg(deconv_kernel):
        """Get configurations for deconv layers."""
        if deconv_kernel == 4:
            padding = 1
            output_padding = 0
        elif deconv_kernel == 3:
            padding = 1
            output_padding = 1
        elif deconv_kernel == 2:
            padding = 0
            output_padding = 0
        else:
            raise ValueError(f'Not supported num_kernels ({deconv_kernel}).')

        return deconv_kernel, padding, output_padding

    def get_loss(self, outputs, targets, masks, joints):
        """Calculate bottom-up keypoint loss.

        Note:
            - batch_size: N
            - num_keypoints: K
            - num_outputs: O
            - heatmaps height: H
            - heatmaps weight: W

        Args:
            outputs (list(torch.Tensor[N,K,H,W])): Multi-scale output heatmaps.
            targets (List(torch.Tensor[N,K,H,W])): Multi-scale target heatmaps.
            masks (List(torch.Tensor[N,H,W])): Masks of multi-scale target
                heatmaps
            joints (List(torch.Tensor[N,M,K,2])): Joints of multi-scale target
                heatmaps for ae loss
        """

        losses = dict()

        heatmaps_losses, push_losses, pull_losses = self.loss(
            outputs, targets, masks, joints)

        for idx in range(len(targets)):
            if heatmaps_losses[idx] is not None:
                heatmaps_loss = heatmaps_losses[idx].mean(dim=0)
                if 'heatmap_loss' not in losses:
                    losses['heatmap_loss'] = heatmaps_loss
                else:
                    losses['heatmap_loss'] += heatmaps_loss
            if push_losses[idx] is not None:
                push_loss = push_losses[idx].mean(dim=0)
                if 'push_loss' not in losses:
                    losses['push_loss'] = push_loss
                else:
                    losses['push_loss'] += push_loss
            if pull_losses[idx] is not None:
                pull_loss = pull_losses[idx].mean(dim=0)
                if 'pull_loss' not in losses:
                    losses['pull_loss'] = pull_loss
                else:
                    losses['pull_loss'] += pull_loss

        return losses

    def forward(self, x):
        """Forward function."""
        if isinstance(x, list):
            x = x[0]

        final_outputs = []
        y = self.final_layers[0](x)
        final_outputs.append(y)

        for i in range(self.num_deconvs):
            if self.cat_output[i]:
                x = torch.cat((x, y), 1)

            x = self.deconv_layers[i](x)
            y = self.final_layers[i + 1](x)
            final_outputs.append(y)

        return final_outputs

    def init_weights(self):
        """Initialize model weights."""
        for _, m in self.deconv_layers.named_modules():
            if isinstance(m, nn.ConvTranspose2d):
                normal_init(m, std=0.001)
            elif isinstance(m, nn.BatchNorm2d):
                constant_init(m, 1)
        for _, m in self.final_layers.named_modules():
            if isinstance(m, nn.Conv2d):
                normal_init(m, std=0.001, bias=0)