File size: 4,462 Bytes
67ddbf8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f5651ba
67ddbf8
 
 
 
 
 
 
 
 
f5651ba
 
 
 
67ddbf8
 
 
 
 
 
 
 
 
 
 
 
 
 
f5651ba
 
 
67ddbf8
 
 
f5651ba
67ddbf8
 
 
 
 
 
 
 
f5651ba
67ddbf8
 
 
 
f5651ba
 
 
67ddbf8
 
 
 
f5651ba
 
 
 
 
67ddbf8
7a05005
67ddbf8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f5651ba
 
 
67ddbf8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7a05005
 
67ddbf8
 
 
 
f5651ba
 
 
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
# Copyright (c) 2024 Bytedance Ltd. and/or its affiliates
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import sys
import tqdm
from eval.syncnet import SyncNetEval
from eval.syncnet_detect import SyncNetDetector
from eval.eval_sync_conf import syncnet_eval
import torch
import subprocess
import shutil
from multiprocessing import Process

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from config import MODELS_DIR


paths = []


def gather_paths(input_dir, output_dir):
    # os.makedirs(output_dir, exist_ok=True)

    for video in tqdm.tqdm(sorted(os.listdir(input_dir))):
        if video.endswith(".mp4"):
            video_input = os.path.join(input_dir, video)
            video_output = os.path.join(output_dir, video)
            if os.path.isfile(video_output):
                continue
            paths.append((video_input, video_output))
        elif os.path.isdir(os.path.join(input_dir, video)):
            gather_paths(
                os.path.join(input_dir, video), os.path.join(output_dir, video)
            )


def adjust_offset(video_input: str, video_output: str, av_offset: int, fps: int = 25):
    command = f"ffmpeg -loglevel error -y -i {video_input} -itsoffset {av_offset / fps} -i {video_input} -map 0:v -map 1:a -c copy -q:v 0 -q:a 0 {video_output}"
    subprocess.run(command, shell=True)


def func(sync_conf_threshold, paths, device_id, process_temp_dir):
    os.makedirs(process_temp_dir, exist_ok=True)
    device = f"cuda:{device_id}"

    syncnet = SyncNetEval(device=device)
    syncnet.loadParameters(f"{MODELS_DIR}/auxiliary/syncnet_v2.model")

    detect_results_dir = os.path.join(process_temp_dir, "detect_results")
    syncnet_eval_results_dir = os.path.join(process_temp_dir, "syncnet_eval_results")

    syncnet_detector = SyncNetDetector(
        device=device, detect_results_dir=detect_results_dir
    )

    for video_input, video_output in paths:
        try:
            av_offset, conf = syncnet_eval(
                syncnet,
                syncnet_detector,
                video_input,
                syncnet_eval_results_dir,
                detect_results_dir,
            )

            if conf >= sync_conf_threshold and abs(av_offset) <= 6:
                os.makedirs(os.path.dirname(video_output), exist_ok=True)
                if av_offset == 0:
                    shutil.copy(video_input, video_output)
                else:
                    adjust_offset(video_input, video_output, av_offset)
        except Exception as e:
            print(e)


def split(a, n):
    k, m = divmod(len(a), n)
    return (a[i * k + min(i, m) : (i + 1) * k + min(i + 1, m)] for i in range(n))


def sync_av_multi_gpus(
    input_dir, output_dir, temp_dir, num_workers, sync_conf_threshold
):
    gather_paths(input_dir, output_dir)
    num_devices = torch.cuda.device_count()
    if num_devices == 0:
        raise RuntimeError("No GPUs found")
    split_paths = list(split(paths, num_workers * num_devices))
    processes = []

    for i in range(num_devices):
        for j in range(num_workers):
            process_index = i * num_workers + j
            process = Process(
                target=func,
                args=(
                    sync_conf_threshold,
                    split_paths[process_index],
                    i,
                    os.path.join(temp_dir, f"process_{process_index}"),
                ),
            )
            process.start()
            processes.append(process)

    for process in processes:
        process.join()


if __name__ == "__main__":
    input_dir = "/mnt/bn/maliva-gen-ai-v2/chunyu.li/VoxCeleb2/affine_transformed"
    output_dir = "/mnt/bn/maliva-gen-ai-v2/chunyu.li/VoxCeleb2/av_synced"
    temp_dir = "temp"
    num_workers = 20  # How many processes per device
    sync_conf_threshold = 3

    sync_av_multi_gpus(
        input_dir, output_dir, temp_dir, num_workers, sync_conf_threshold
    )