File size: 7,136 Bytes
d7e58f0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import numpy as np
import torch
from smplx import FLAME as _FLAME
from smplx import FLAMELayer as _FLAMELayer

from detrsmpl.core.conventions.keypoints_mapping import (
    convert_kps,
    get_keypoint_num,
)


class FLAME(_FLAME):
    """Extension of the official FLAME implementation."""
    head_pose_keys = {'global_orient', 'jaw_pose'}
    full_pose_keys = {
        'global_orient', 'neck_pose', 'jaw_pose', 'leye_pose', 'reye_pose'
    }

    NUM_VERTS = 5023
    NUM_FACES = 9976

    def __init__(self,
                 *args,
                 keypoint_src: str = 'flame',
                 keypoint_dst: str = 'human_data',
                 keypoint_approximate: bool = False,
                 **kwargs):
        """
        Args:
            *args: extra arguments for FLAME initialization.
            keypoint_src: source convention of keypoints. This convention
                is used for keypoints obtained from joint regressors.
                Keypoints then undergo conversion into keypoint_dst
                convention.
            keypoint_dst: destination convention of keypoints. This convention
                is used for keypoints in the output.
            keypoint_approximate: whether to use approximate matching in
                convention conversion for keypoints.
            **kwargs: extra keyword arguments for FLAME initialization.

        Returns:
            None
        """
        super(FLAME, self).__init__(*args, **kwargs)
        self.keypoint_src = keypoint_src
        self.keypoint_dst = keypoint_dst
        self.keypoint_approximate = keypoint_approximate

        self.num_verts = self.get_num_verts()
        self.num_faces = self.get_num_faces()
        self.num_joints = get_keypoint_num(convention=self.keypoint_dst)

    def forward(self,
                *args,
                return_verts: bool = True,
                return_full_pose: bool = False,
                **kwargs) -> dict:
        """Forward function.

        Args:
            *args: extra arguments for FLAME
            return_verts: whether to return vertices
            return_full_pose: whether to return full pose parameters
            **kwargs: extra arguments for FLAME

        Returns:
            output: contains output parameters and attributes
        """
        flame_output = super(FLAME, self).forward(*args, **kwargs)
        joints = flame_output.joints
        joints, joint_mask = convert_kps(joints,
                                         src=self.keypoint_src,
                                         dst=self.keypoint_dst,
                                         approximate=self.keypoint_approximate)
        if isinstance(joint_mask, np.ndarray):
            joint_mask = torch.tensor(joint_mask,
                                      dtype=torch.uint8,
                                      device=joints.device)

        batch_size = joints.shape[0]
        joint_mask = joint_mask.reshape(1, -1).expand(batch_size, -1)

        output = dict(global_orient=flame_output.global_orient,
                      neck_pose=flame_output.neck_pose,
                      jaw_pose=flame_output.jaw_pose,
                      joints=joints,
                      joint_mask=joint_mask,
                      keypoints=torch.cat([joints, joint_mask[:, :, None]],
                                          dim=-1),
                      betas=flame_output.betas,
                      expression=flame_output.expression)

        if return_verts:
            output['vertices'] = flame_output.vertices
        if return_full_pose:
            output['full_pose'] = flame_output.full_pose

        return output


class FLAMELayer(_FLAMELayer):
    """Extension of the official FLAME implementation."""
    head_pose_keys = {'global_orient', 'jaw_pose'}
    full_pose_keys = {
        'global_orient', 'neck_pose', 'jaw_pose', 'leye_pose', 'reye_pose'
    }

    NUM_VERTS = 5023
    NUM_FACES = 9976

    def __init__(self,
                 *args,
                 keypoint_src: str = 'flame',
                 keypoint_dst: str = 'human_data',
                 keypoint_approximate: bool = False,
                 **kwargs):
        """
        Args:
            *args: extra arguments for FLAME initialization.
            keypoint_src: source convention of keypoints. This convention
                is used for keypoints obtained from joint regressors.
                Keypoints then undergo conversion into keypoint_dst
                convention.
            keypoint_dst: destination convention of keypoints. This convention
                is used for keypoints in the output.
            keypoint_approximate: whether to use approximate matching in
                convention conversion for keypoints.
            **kwargs: extra keyword arguments for FLAME initialization.

        Returns:
            None
        """
        super(FLAMELayer, self).__init__(*args, **kwargs)
        self.keypoint_src = keypoint_src
        self.keypoint_dst = keypoint_dst
        self.keypoint_approximate = keypoint_approximate

        self.num_verts = self.get_num_verts()
        self.num_faces = self.get_num_faces()
        self.num_joints = get_keypoint_num(convention=self.keypoint_dst)

    def forward(self,
                *args,
                return_verts: bool = True,
                return_full_pose: bool = False,
                **kwargs) -> dict:
        """Forward function.

        Args:
            *args: extra arguments for FLAME
            return_verts: whether to return vertices
            return_full_pose: whether to return full pose parameters
            **kwargs: extra arguments for FLAME

        Returns:
            output: contains output parameters and attributes
        """
        flame_output = super(FLAMELayer, self).forward(*args, **kwargs)
        joints = flame_output.joints
        joints, joint_mask = convert_kps(joints,
                                         src=self.keypoint_src,
                                         dst=self.keypoint_dst,
                                         approximate=self.keypoint_approximate)
        if isinstance(joint_mask, np.ndarray):
            joint_mask = torch.tensor(joint_mask,
                                      dtype=torch.uint8,
                                      device=joints.device)

        batch_size = joints.shape[0]
        joint_mask = joint_mask.reshape(1, -1).expand(batch_size, -1)

        output = dict(global_orient=flame_output.global_orient,
                      neck_pose=flame_output.neck_pose,
                      jaw_pose=flame_output.jaw_pose,
                      joints=joints,
                      joint_mask=joint_mask,
                      keypoints=torch.cat([joints, joint_mask[:, :, None]],
                                          dim=-1),
                      betas=flame_output.betas,
                      expression=flame_output.expression)

        if return_verts:
            output['vertices'] = flame_output.vertices
        if return_full_pose:
            output['full_pose'] = flame_output.full_pose

        return output