File size: 6,261 Bytes
50ee618
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Generate a dancing skeleton video from any audio file.

Usage:
    python generate.py --audio song.wav --out dance.mp4
"""

from __future__ import annotations

import argparse
import os
import subprocess

import librosa
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import matplotlib.animation as animation
import numpy as np
import soundfile as sf
import torch
from tqdm import tqdm

from audio_features import AUDIO_SR, POSE_FPS, audio_to_features
from inference import generate_poses, load_checkpoint, resolve_checkpoint

SKELETON_EDGES = [
    (0, 1), (1, 2), (2, 3), (3, 7), (0, 4), (4, 5), (5, 6), (6, 8), (9, 10),
    (11, 12), (11, 13), (13, 15), (12, 14), (14, 16),
    (11, 23), (12, 24), (23, 24), (23, 25), (24, 26), (25, 27), (26, 28),
    (27, 29), (28, 30), (29, 31), (30, 32),
]
FACE = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
LEFT = {11, 13, 15, 23, 25, 27, 29, 31}
RIGHT = {12, 14, 16, 24, 26, 28, 30, 32}


def load_audio(path: str) -> np.ndarray:
    """Load any audio file as mono float32 at 32 kHz."""
    data, sr = sf.read(path, dtype="float32", always_2d=True)
    mono = data.mean(axis=1)
    if sr != AUDIO_SR:
        mono = librosa.resample(mono, orig_sr=sr, target_sr=AUDIO_SR)
    return mono.astype(np.float32)


def _segment_color(a: int, b: int) -> str:
    if a in FACE or b in FACE:
        return "#a0c4ff"
    if a in LEFT and b in LEFT:
        return "#ffd6a5"
    if a in RIGHT and b in RIGHT:
        return "#ffadad"
    return "#b9fbc0"


def render_skeleton_video(poses_xyz: np.ndarray, title: str, tmp_path: str) -> None:
    """Render a silent skeleton animation (720×1080)."""
    T = poses_xyz.shape[0]
    vel = np.diff(poses_xyz, axis=0)
    energy = np.concatenate([[0], np.linalg.norm(vel, axis=-1).mean(axis=-1)])
    energy = energy / (energy.max() + 1e-6)

    xmin = poses_xyz[:, :, 0].min() - 0.15
    xmax = poses_xyz[:, :, 0].max() + 0.15
    ymin = poses_xyz[:, :, 1].min() - 0.1
    ymax = poses_xyz[:, :, 1].max() + 0.1

    fig = plt.figure(figsize=(5.4, 8), facecolor="#0d1117")
    ax = fig.add_axes([0.05, 0.08, 0.90, 0.86])
    ax.set_facecolor("#0d1117")
    ax.set_xlim(xmin, xmax)
    ax.set_ylim(-ymax, -ymin)
    ax.set_aspect("equal")
    ax.axis("off")
    fig.text(0.5, 0.97, title, color="#e6edf3", fontsize=11,
             ha="center", va="top", fontweight="bold")

    bar_ax = fig.add_axes([0.05, 0.02, 0.90, 0.04])
    bar_ax.set_xlim(0, T)
    bar_ax.set_ylim(0, 1)
    bar_ax.axis("off")
    progress_bar = bar_ax.barh(0.5, 0, height=0.8, color="#3fb950", left=0)
    time_txt = bar_ax.text(T * 0.5, 0.5, "0.0 s", color="white",
                           fontsize=7, ha="center", va="center")

    scat = ax.scatter([], [], s=18, zorder=4)
    lines = [ax.plot([], [], lw=2.2, solid_capstyle="round")[0] for _ in SKELETON_EDGES]

    def init():
        scat.set_offsets(np.empty((0, 2)))
        for line in lines:
            line.set_data([], [])
        return [scat, *lines]

    def update(t):
        kpts = poses_xyz[t]
        xs, ys = kpts[:, 0], -kpts[:, 1]
        cols = []
        for i in range(33):
            if i in FACE:
                cols.append("#a0c4ff")
            elif i in LEFT:
                cols.append("#ffd6a5")
            elif i in RIGHT:
                cols.append("#ffadad")
            else:
                cols.append("#b9fbc0")
        scat.set_offsets(np.c_[xs, ys])
        scat.set_color(cols)
        for idx, (a, b) in enumerate(SKELETON_EDGES):
            lines[idx].set_data([xs[a], xs[b]], [ys[a], ys[b]])
            lines[idx].set_color(_segment_color(a, b))
            lines[idx].set_alpha(0.85 + 0.15 * energy[t])
        progress_bar[0].set_width(t + 1)
        time_txt.set_text(f"{t / POSE_FPS:.1f} s / {T / POSE_FPS:.1f} s")
        return [scat, *lines, progress_bar[0], time_txt]

    ani = animation.FuncAnimation(
        fig, update, frames=T, init_func=init,
        interval=1000 / POSE_FPS, blit=True,
    )
    writer = animation.FFMpegWriter(
        fps=POSE_FPS, bitrate=2000,
        extra_args=["-vcodec", "libx264", "-pix_fmt", "yuv420p"],
    )
    ani.save(tmp_path, writer=writer, dpi=150)
    plt.close(fig)


def mux_audio(video_path: str, audio_path: str, out_path: str, duration: float) -> None:
    """Combine silent video with the original audio track."""
    subprocess.run([
        "ffmpeg", "-y",
        "-i", video_path,
        "-i", audio_path,
        "-c:v", "copy",
        "-c:a", "aac", "-b:a", "192k",
        "-t", str(duration),
        "-shortest",
        out_path,
    ], check=True, capture_output=True)


def main() -> None:
    parser = argparse.ArgumentParser(description="Music → dance skeleton video")
    parser.add_argument("--audio", required=True, help="Input audio file")
    parser.add_argument("--checkpoint", default=None, help="Path to .pt weights")
    parser.add_argument("--out", default="dance.mp4", help="Output video path")
    args = parser.parse_args()

    os.makedirs(os.path.dirname(args.out) or ".", exist_ok=True)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    ckpt_path = resolve_checkpoint(args.checkpoint)

    print(f"Device: {device}")
    print(f"Loading audio: {args.audio}")
    waveform = load_audio(args.audio)
    duration = len(waveform) / AUDIO_SR
    print(f"  Duration: {duration:.1f}s")

    print("Extracting audio features…")
    audio_feat = audio_to_features(waveform)
    print(f"  {audio_feat.shape[0]} frames @ {POSE_FPS} fps")

    print(f"Loading model: {ckpt_path}")
    model, ckpt = load_checkpoint(ckpt_path, device)
    print(f"  Epoch {ckpt['epoch']}")

    print("Generating poses…")
    poses_xyz = generate_poses(
        model, audio_feat,
        ckpt["x_mean"], ckpt["x_std"],
        ckpt["y_mean"], ckpt["y_std"],
        device,
    )

    title = os.path.splitext(os.path.basename(args.audio))[0]
    tmp = args.out.replace(".mp4", "_silent.mp4")
    print("Rendering video…")
    render_skeleton_video(poses_xyz, title=title, tmp_path=tmp)

    print("Mixing audio…")
    mux_audio(tmp, args.audio, args.out, duration)
    os.remove(tmp)
    print(f"Done → {args.out}")


if __name__ == "__main__":
    main()