File size: 6,685 Bytes
347ea73
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import math

import numpy as np
import torch
import torch.nn as nn
import torchvision.models.resnet as resnet


def rot6d_to_rotmat(x):
    """Convert 6D rotation representation to 3x3 rotation matrix.
    Based on Zhou et al., "On the Continuity of Rotation Representations in Neural Networks", CVPR 2019
    Input:
        (B,6) Batch of 6-D rotation representations
    Output:
        (B,3,3) Batch of corresponding rotation matrices
    """

    x = x.view(-1, 3, 2)
    a1 = x[:, :, 0]
    a2 = x[:, :, 1]
    b1 = nn.functional.normalize(a1)
    b2 = nn.functional.normalize(
        a2 - torch.einsum("bi,bi->b", b1, a2).unsqueeze(-1) * b1
    )

    b3 = torch.cross(b1, b2)

    return torch.stack((b1, b2, b3), dim=-1)


class Bottleneck(nn.Module):
    """Redefinition of Bottleneck residual block
    Adapted from the official PyTorch implementation
    """

    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(
            planes, planes, kernel_size=3, stride=stride, padding=1, bias=False
        )
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * 4)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out


class HMR(nn.Module):
    """SMPL Iterative Regressor with ResNet50 backbone"""

    def __init__(self, block, layers, smpl_mean_params):
        self.inplanes = 64
        super(HMR, self).__init__()
        self.n_shape = 10
        self.n_cam = 3
        self.n_joints = 24
        npose = self.n_joints * 6
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.avgpool = nn.AvgPool2d(7, stride=1)
        self.fc1 = nn.Linear(512 * block.expansion + npose + self.n_shape + self.n_cam, 1024)
        self.drop1 = nn.Dropout()
        self.fc2 = nn.Linear(1024, 1024)
        self.drop2 = nn.Dropout()
        self.decpose = nn.Linear(1024, npose)
        self.decshape = nn.Linear(1024, self.n_shape)
        self.deccam = nn.Linear(1024, self.n_cam)
        nn.init.xavier_uniform_(self.decpose.weight, gain=0.01)
        nn.init.xavier_uniform_(self.decshape.weight, gain=0.01)
        nn.init.xavier_uniform_(self.deccam.weight, gain=0.01)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2.0 / n))
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

        mean_params = np.load(smpl_mean_params)
        init_pose = torch.from_numpy(mean_params["pose"][:]).unsqueeze(0)
        init_shape = torch.from_numpy(
            mean_params["shape"][:].astype("float32")
        ).unsqueeze(0)
        init_cam = torch.from_numpy(mean_params["cam"]).unsqueeze(0)
        self.register_buffer("init_pose", init_pose)
        self.register_buffer("init_shape", init_shape)
        self.register_buffer("init_cam", init_cam)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(
                    self.inplanes,
                    planes * block.expansion,
                    kernel_size=1,
                    stride=stride,
                    bias=False,
                ),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def forward(self, x, init_pose=None, init_shape=None, init_cam=None, n_iter=3):

        batch_size = x.shape[0]

        if init_pose is None:
            init_pose = self.init_pose.expand(batch_size, -1)
        if init_shape is None:
            init_shape = self.init_shape.expand(batch_size, -1)
        if init_cam is None:
            init_cam = self.init_cam.expand(batch_size, -1)

        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x1 = self.layer1(x)
        x2 = self.layer2(x1)
        x3 = self.layer3(x2)
        x4 = self.layer4(x3)

        xf = self.avgpool(x4)
        xf = xf.view(xf.size(0), -1)

        pred_pose = init_pose
        pred_shape = init_shape
        pred_cam = init_cam
        for _ in range(n_iter):
            xc = torch.cat([xf, pred_pose, pred_shape, pred_cam], 1)
            xc = self.fc1(xc)
            xc = self.drop1(xc)
            xc = self.fc2(xc)
            xc = self.drop2(xc)
            pred_pose = self.decpose(xc) + pred_pose
            pred_shape = self.decshape(xc) + pred_shape
            pred_cam = self.deccam(xc) + pred_cam

        pred_rotmat = rot6d_to_rotmat(pred_pose).view(batch_size, self.n_joints, 3, 3)

        return pred_rotmat, pred_shape, pred_cam


def hmr(smpl_mean_params, pretrained=True, **kwargs):
    """Constructs an HMR model with ResNet50 backbone.
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = HMR(Bottleneck, [3, 4, 6, 3], smpl_mean_params, **kwargs)
    if pretrained:
        resnet_imagenet = resnet.resnet50(pretrained=True)
        model.load_state_dict(resnet_imagenet.state_dict(), strict=False)
    return model