File size: 5,602 Bytes
882f6e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Copyright (c) Meta Platforms, Inc. and affiliates.
All rights reserved.
This source code is licensed under the license found in the
LICENSE file in the root directory of this source tree.
"""

import copy
import glob
import os
import re
import subprocess
from collections import OrderedDict
from typing import Dict, List

import mediapy

import numpy as np

import torch
import torch as th
import torchaudio
from attrdict import AttrDict

from omegaconf import OmegaConf
from tqdm import tqdm
from utils.model_util import get_person_num
from visualize.ca_body.utils.image import linear2displayBatch
from visualize.ca_body.utils.train import load_checkpoint, load_from_config

ffmpeg_header = "ffmpeg -y "  # -hide_banner -loglevel error "


def filter_params(params, ignore_names):
    return OrderedDict(
        [
            (k, v)
            for k, v in params.items()
            if not any([re.match(n, k) is not None for n in ignore_names])
        ]
    )


def call_ffmpeg(command: str) -> None:
    print(command, "-" * 100)
    e = subprocess.call(command, shell=True)
    if e != 0:
        assert False, e


class BodyRenderer(th.nn.Module):
    def __init__(
        self,
        config_base: str,
        render_rgb: bool,
    ):
        super().__init__()
        self.config_base = config_base
        ckpt_path = f"{config_base}/body_dec.ckpt"
        config_path = f"{config_base}/config.yml"
        assets_path = f"{config_base}/static_assets.pt"
        # config
        config = OmegaConf.load(config_path)
        gpu = config.get("gpu", 0)
        self.device = th.device(f"cuda:{gpu}")
        # assets
        static_assets = AttrDict(torch.load(assets_path))
        # build model
        self.model = load_from_config(config.model, assets=static_assets).to(
            self.device
        )
        self.model.cal_enabled = False
        self.model.pixel_cal_enabled = False
        self.model.learn_blur_enabled = False
        self.render_rgb = render_rgb
        if not self.render_rgb:
            self.model.rendering_enabled = None
        # load model checkpoints
        print("loading...", ckpt_path)
        load_checkpoint(
            ckpt_path,
            modules={"model": self.model},
            ignore_names={"model": ["lbs_fn.*"]},
        )
        self.model.eval()
        self.model.to(self.device)
        # load default parameters for renderer
        person = get_person_num(config_path)
        self.default_inputs = th.load(f"assets/render_defaults_{person}.pth")

    def _write_video_stream(
        self, motion: np.ndarray, face: np.ndarray, save_name: str
    ) -> None:
        out = self._render_loop(motion, face)
        mediapy.write_video(save_name, out, fps=30)

    def _render_loop(self, body_pose: np.ndarray, face: np.ndarray) -> List[np.ndarray]:
        all_rgb = []
        default_inputs_copy = copy.deepcopy(self.default_inputs)
        for b in tqdm(range(len(body_pose))):
            B = default_inputs_copy["K"].shape[0]
            default_inputs_copy["lbs_motion"] = (
                th.tensor(body_pose[b : b + 1, :], device=self.device, dtype=th.float)
                .tile(B, 1)
                .to(self.device)
            )
            geom = (
                self.model.lbs_fn.lbs_fn(
                    default_inputs_copy["lbs_motion"],
                    self.model.lbs_fn.lbs_scale.unsqueeze(0).tile(B, 1),
                    self.model.lbs_fn.lbs_template_verts.unsqueeze(0).tile(B, 1, 1),
                )
                * self.model.lbs_fn.global_scaling
            )
            default_inputs_copy["geom"] = geom
            face_codes = (
                th.from_numpy(face).float().cuda() if not th.is_tensor(face) else face
            )
            curr_face = th.tile(face_codes[b : b + 1, ...], (2, 1))
            default_inputs_copy["face_embs"] = curr_face
            preds = self.model(**default_inputs_copy)
            rgb0 = linear2displayBatch(preds["rgb"])[0]
            rgb1 = linear2displayBatch(preds["rgb"])[1]
            rgb = th.cat((rgb0, rgb1), axis=-1).permute(1, 2, 0)
            rgb = rgb.clip(0, 255).to(th.uint8)
            all_rgb.append(rgb.contiguous().detach().byte().cpu().numpy())
        return all_rgb

    def render_full_video(
        self,
        data_block: Dict[str, np.ndarray],
        animation_save_path: str,
        audio_sr: int = None,
        render_gt: bool = False,
    ) -> None:
        tag = os.path.basename(os.path.dirname(animation_save_path))
        save_name = os.path.splitext(os.path.basename(animation_save_path))[0]
        save_name = f"{tag}_{save_name}"
        torchaudio.save(
            f"/tmp/audio_{save_name}.wav",
            torch.tensor(data_block["audio"]),
            audio_sr,
        )
        if render_gt:
            tag = "gt"
            self._write_video_stream(
                data_block["gt_body"],
                data_block["gt_face"],
                f"/tmp/{tag}_{save_name}.mp4",
            )
        else:
            tag = "pred"
            self._write_video_stream(
                data_block["body_motion"],
                data_block["face_motion"],
                f"/tmp/{tag}_{save_name}.mp4",
            )
        command = f"{ffmpeg_header} -i /tmp/{tag}_{save_name}.mp4 -i /tmp/audio_{save_name}.wav -c:v copy -map 0:v:0 -map 1:a:0 -c:a aac -b:a 192k -pix_fmt yuva420p {animation_save_path}_{tag}.mp4"
        call_ffmpeg(command)
        subprocess.call(
            f"rm /tmp/audio_{save_name}.wav && rm /tmp/{tag}_{save_name}.mp4",
            shell=True,
        )