File size: 6,502 Bytes
b03a8f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import numpy as np
import torch
import smplx

MASK_DICT = {
    "local_upper": [
        False, False, False, True, False, False, True, False, False, True,
        False, False, True, True, True, True, True, True, True, True,
        True, True, False, False, False, True, True, True, True, True,
        True, True, True, True, True, True, True, True, True, True,
        True, True, True, True, True, True, True, True, True, True,
        True, True, True, True, True
    ],
    "local_full": [False] + [True]*54
}

def select_with_mask(motion: np.ndarray, mask: list[bool]) -> np.ndarray:
    mask_arr = np.array(mask, dtype=bool)
    j = len(mask_arr)
    c_channels = motion.shape[-1] // j
    new_shape = motion.shape[:-1] + (j, c_channels)
    motion = motion.reshape(new_shape)
    selected_motion = motion[..., mask_arr, :]
    final_shape = selected_motion.shape[:-2] + (selected_motion.shape[-2]*selected_motion.shape[-1],)
    selected_motion = selected_motion.reshape(final_shape)
    return selected_motion

def recover_from_mask(selected_motion: np.ndarray, mask: list[bool]) -> np.ndarray:
    mask_arr = np.array(mask, dtype=bool)
    j = len(mask_arr)
    # Infer c_channels from selected_motion's last dimension and sum(mask)
    c_channels = selected_motion.shape[-1] // mask_arr.sum()
    new_shape = selected_motion.shape[:-1] + (mask_arr.sum(), c_channels)
    selected_motion = selected_motion.reshape(new_shape)
    out_shape = selected_motion.shape[:-2] + (j, c_channels)
    recovered = np.zeros(out_shape, dtype=selected_motion.dtype)
    recovered[..., mask_arr, :] = selected_motion
    final_shape = recovered.shape[:-2] + (j * c_channels,)
    recovered = recovered.reshape(final_shape)
    return recovered

def select_with_mask_ts(motion: torch.Tensor, mask: list[bool]) -> torch.Tensor:
    mask_arr = torch.tensor(mask, dtype=torch.bool, device=motion.device)
    j = len(mask_arr)
    c_channels = motion.shape[-1] // j
    new_shape = motion.shape[:-1] + (j, c_channels)
    motion = motion.reshape(new_shape)
    selected_motion = motion[..., mask_arr, :]
    final_shape = selected_motion.shape[:-2] + (selected_motion.shape[-2]*selected_motion.shape[-1],)
    selected_motion = selected_motion.reshape(final_shape)
    return selected_motion

def recover_from_mask_ts(selected_motion: torch.Tensor, mask: list[bool]) -> torch.Tensor:
    device = selected_motion.device
    dtype = selected_motion.dtype
    mask_arr = torch.tensor(mask, dtype=torch.bool, device=device)
    j = len(mask_arr)
    sum_mask = mask_arr.sum().item()
    c_channels = selected_motion.shape[-1] // sum_mask
    new_shape = selected_motion.shape[:-1] + (sum_mask, c_channels)
    selected_motion = selected_motion.reshape(new_shape)
    out_shape = list(selected_motion.shape[:-2]) + [j, c_channels]
    recovered = torch.zeros(out_shape, dtype=dtype, device=device)
    recovered[..., mask_arr, :] = selected_motion
    final_shape = list(recovered.shape[:-2]) + [j * c_channels]
    recovered = recovered.reshape(final_shape)
    return recovered

def time_upsample_numpy(data: np.ndarray, k: int) -> np.ndarray:
    # data: (..., t, c)
    # output: (..., k*t, c)
    if k == 1:
        return data.copy()
    shape = data.shape
    t = shape[-2]
    c = shape[-1]
    # original and new time indices
    original_t = np.arange(t)
    new_t = np.linspace(0, t - 1, k * t)

    # reshape to (M, c, t)
    reshaped = data.reshape(-1, t, c).transpose(0, 2, 1)
    M = reshaped.shape[0]
    reshaped = reshaped.reshape(M * c, t)

    # find interpolation indices
    idx = np.searchsorted(original_t, new_t, side='right') - 1
    idx = np.clip(idx, 0, t - 2)
    idx1 = idx + 1

    x0 = original_t[idx]
    x1 = original_t[idx1]
    w = (new_t - x0) / (x1 - x0)

    f0 = reshaped[:, idx]
    f1 = reshaped[:, idx1]

    out = f0 + (f1 - f0) * w
    out = out.reshape(M, c, k * t).transpose(0, 2, 1)
    final_shape = shape[:-2] + (k * t, c)
    return out.reshape(final_shape)

def beat_format_save(
    save_path: str,
    motion_data: np.ndarray,
    mask: list[bool] = None,
    betas: np.ndarray = None,
    expressions: np.ndarray = None,
    trans: np.ndarray = None,
    upsample: int = None,
):
    if betas is None:
        betas = np.zeros((motion_data.shape[0], 300), dtype=motion_data.dtype)
    if expressions is None:
        expressions = np.zeros((motion_data.shape[0], 100), dtype=motion_data.dtype)
    if trans is None:
        smplx_model = smplx.create(
            "./emage_evaltools/smplx_models/",
            model_type='smplx',
            gender='NEUTRAL_2020',
            use_face_contour=False,
            num_betas=300,
            num_expression_coeffs=100,
            ext='npz',
            use_pca=False
        ).eval()
        betas_ts = torch.from_numpy(betas[0:1]).float()
        output = smplx_model(
            betas=betas_ts,
            transl=torch.zeros(1, 3),
            expression=torch.zeros(1, 100),
            jaw_pose=torch.zeros(1, 3),
            global_orient=torch.zeros(1, 3),
            body_pose=torch.zeros(1, 63),
            left_hand_pose=torch.zeros(1, 45),
            right_hand_pose=torch.zeros(1, 45),
            return_joints=True,
            leye_pose=torch.zeros(1, 3),
            reye_pose=torch.zeros(1, 3)
        )
        trans = (output["joints"][:, 10, :] + output["joints"][:, 11, :]) / 2
        # print(trans)
        trans = -trans.repeat(motion_data.shape[0], 1).numpy()

    if mask is not None:
        motion_data = recover_from_mask(motion_data, mask)

    if upsample is not None and upsample > 1:
        motion_data = time_upsample_numpy(motion_data, upsample)
        betas = time_upsample_numpy(betas, upsample)
        expressions = time_upsample_numpy(expressions, upsample)
        trans = time_upsample_numpy(trans, upsample)

    np.savez(
        save_path,
        betas=betas[0],
        poses=motion_data,
        expressions=expressions,
        trans=trans,
        model='smplx2020',
        gender='neutral',
        mocap_frame_rate=30
    )

def beat_format_load(load_path: str, mask: list[bool] = None):
    data = np.load(load_path, allow_pickle=True)
    poses = data['poses']
    betas = data['betas']
    expressions = data['expressions']
    trans = data['trans']

    if mask is not None:
        poses = select_with_mask(poses, mask)

    return {
        "poses": poses,
        "betas": betas,
        "expressions": expressions,
        "trans": trans
    }