Hugging Face's logo

Datasets:
xingzhaohu
/
temp_dataset

Dataset card Files
xet
Community
temp_dataset
File size: 4,582 Bytes 
4b3a024
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import os
import subprocess

import cv2
import numpy as np

from tqdm import tqdm

from speakervid_data_talkinghead import SpeakerVidTalkingDataset

S3_ENDPOINT_URL = "https://t3.storage.dev"
AWS_ACCESS_KEY_ID = "tid_cqKPLHboixMUUQxq_ImANLFwrehWmWZHlEaPZXzXNbKxf_fugg"
AWS_SECRET_ACCESS_KEY = "tsec_CXLclBpmOD2blVqdL+smpI52cOxQiXs-pH-INnfU6yfhc1MAajUTpI7xWO+5YAyLwyXjpq"


def _visualize_face_mask(video, face_mask, out_path, fps=25, alpha=0.5) -> None:
    frames = (
        ((video + 1.0) * 127.5)
        .clamp(0, 255)
        .byte()
        .permute(0, 2, 3, 1)
        .cpu()
        .numpy()
    )
    mask = face_mask.squeeze(1).cpu().numpy()
    h, w = frames.shape[1], frames.shape[2]
    if mask.shape[1] != h or mask.shape[2] != w:
        resized = np.zeros((mask.shape[0], h, w), dtype=np.float32)
        for i in range(mask.shape[0]):
            resized[i] = cv2.resize(mask[i], (w, h), interpolation=cv2.INTER_NEAREST)
        mask = resized

    writer = cv2.VideoWriter(
        out_path, cv2.VideoWriter_fourcc(*"mp4v"), fps, (w, h)
    )
    for i, frame in enumerate(frames):
        overlay = np.zeros_like(frame)
        overlay[:, :, 2] = (mask[i] > 0.5).astype(np.uint8) * 255
        blended = cv2.addWeighted(frame, 1.0, overlay, alpha, 0.0)
        writer.write(cv2.cvtColor(blended, cv2.COLOR_RGB2BGR))
    writer.release()
def main() -> None:
    config = {
        "jsonl_path": "/mnt/nfs/datasets/SpeakerVid-5M/metadb_code/talking_top15_syncc.jsonl",
        "existing_tsv_path": "/mnt/nfs/datasets/SpeakerVid-5M/dataprocess_code/output_top15/existing.tsv",
        "audio_feature_model_id": "facebook/wav2vec2-base-960h",
        "filter_enabled": True,
        "sync_d_threshold": 10,  # Sync-D (lower is better)
        "sync_c_threshold": 6.5,  # Sync-C (higher is better)
        "debug_audio": False,
        "use_placeholder_caption": True
    }

    # Optional: override creds via env or config if needed.
    # if os.getenv("AWS_ACCESS_KEY_ID") and os.getenv("AWS_SECRET_ACCESS_KEY"):
    config["aws_access_key_id"] = AWS_ACCESS_KEY_ID # os.getenv("AWS_ACCESS_KEY_ID")
    config["aws_secret_access_key"] =  AWS_SECRET_ACCESS_KEY #os.getenv("AWS_SECRET_ACCESS_KEY")

    dataset = SpeakerVidTalkingDataset(config=config)

    out_dir = os.path.join(os.getcwd(), "visual_tmp")
    os.makedirs(out_dir, exist_ok=True)

    for idx in tqdm(range(min(50, len(dataset)))):
        sample = dataset[idx]
        print("json_name:", sample.get("json_name"))
        print("pixel_values_vid shape:", tuple(sample["pixel_values_vid"].shape))
        print("audio_input_values shape:", tuple(sample["audio_input_values"].shape))
        print("caption:", sample.get("caption_content"))

        # if sample.get("face_mask") is not None:
        #     out_path = os.path.join(out_dir, f"sample_{idx:04d}_mask.mp4")
        #     _visualize_face_mask(sample["pixel_values_vid"], sample["face_mask"], out_path)

        # video = sample["pixel_values_vid"]
        # frames = ((video + 1.0) * 127.5).clamp(0, 255).byte().permute(0, 2, 3, 1).cpu().numpy()
        # out_path = os.path.join(out_dir, f"sample_{idx:04d}.mp4")
        # h, w = frames.shape[1], frames.shape[2]
        # writer = cv2.VideoWriter(out_path, cv2.VideoWriter_fourcc(*"mp4v"), 25, (w, h))
        # for frame in frames:
        #     writer.write(cv2.cvtColor(frame, cv2.COLOR_RGB2BGR))
        # writer.release()

        # audio_clip = sample.get("audio_clip")
        # if audio_clip is None:
        #     continue

        # audio_path = os.path.join(out_dir, f"sample_{idx:04d}.wav")
        # audio_sr = int(sample.get("audio_sample_rate", 16000))
        # audio_clip = np.asarray(audio_clip, dtype=np.float32)

        # try:
        #     import soundfile as sf

        #     sf.write(audio_path, audio_clip, audio_sr)
        # except Exception:
        #     from scipy.io import wavfile

        #     wavfile.write(audio_path, audio_sr, audio_clip)

        # mux_path = os.path.join(out_dir, f"sample_{idx:04d}_av.mp4")
        # subprocess.run(
        #     [
        #         "ffmpeg",
        #         "-y",
        #         "-i",
        #         out_path,
        #         "-i",
        #         audio_path,
        #         "-c:v",
        #         "copy",
        #         "-c:a",
        #         "aac",
        #         "-shortest",
        #         mux_path,
        #     ],
        #     check=True,
        #     stdout=subprocess.DEVNULL,
        #     stderr=subprocess.DEVNULL,
        # )


if __name__ == "__main__":
    main()