File size: 5,569 Bytes
142a1ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import requests
import subprocess
import json
import pandas as pd
import zipfile
import cv2
from pathlib import Path
from tqdm import tqdm

from .video_base import VideoDataset


class SomethingSomethingDataset(VideoDataset):
    """
    Something Something Dataset from https://arxiv.org/abs/1706.04261
    """

    def download(self):
        self.data_root.mkdir(parents=True, exist_ok=True)

        urls = [
            "https://apigwx-aws.qualcomm.com/qsc/public/v1/api/download/software/dataset/AIDataset/Something-Something-V2/20bn-something-something-v2-00",
            "https://apigwx-aws.qualcomm.com/qsc/public/v1/api/download/software/dataset/AIDataset/Something-Something-V2/20bn-something-something-v2-01",
            "https://softwarecenter.qualcomm.com/api/download/software/dataset/AIDataset/Something-Something-V2/20bn-something-something-download-package-labels.zip",
        ]

        for url in urls:
            filename = Path(url).name
            filepath = self.data_root / filename

            print(f"Downloading {filename}...")
            response = requests.get(url, stream=True)
            response.raise_for_status()

            with open(filepath, "wb") as f:
                for chunk in response.iter_content(chunk_size=8192):
                    f.write(chunk)

        # Use shell command to concatenate and extract tar video files
        print("Concatenating and extracting tar files...")
        cmd = f"cd {self.data_root} && cat 20bn-something-something-v2-0? | tar -xvzf -"
        subprocess.run(cmd, shell=True, check=True)
        print(f"Deleting zip files for video data...")
        for zip_file in self.data_root.glob("20bn-something-something-v2-0*"):
            print(f"Deleting {zip_file.name}...")
            zip_file.unlink()

        # Unzip the labels package
        labels_zip_path = (
            self.data_root / "20bn-something-something-download-package-labels.zip"
        )
        if labels_zip_path.exists():
            print(f"Extracting {labels_zip_path.name}...")
            with zipfile.ZipFile(labels_zip_path, "r") as zip_ref:
                zip_ref.extractall(self.data_root)
        print(f"Deleting zip file for labels...")
        labels_zip_path.unlink()

        # Create metadata CSV from labels
        print("Creating metadata CSV file for Something Something Dataset")

        json_files = {
            "training": "labels/train.json",
            "validation": "labels/validation.json",
        }

        records = []
        for split, json_file in json_files.items():
            with open(self.data_root / json_file, "r") as f:
                labels = json.load(f)

            for item in tqdm(labels, desc=f"Creating metadata for {split}"):
                webm_video_path = f"20bn-something-something-v2/{item['id']}.webm"
                mp4_video_path = f"20bn-something-something-v2/{item['id']}.mp4"

                total_videos = len(labels)
                successful_conversions = 0

                if (self.data_root / webm_video_path).exists():
                    # Convert webm to mp4 using ffmpeg
                    input_path = str(self.data_root / webm_video_path)
                    output_path = str(self.data_root / mp4_video_path)
                    cmd = f'ffmpeg -i {input_path} -vf "pad=ceil(iw/2)*2:ceil(ih/2)*2" -c:v libx264 -c:a aac {output_path}'
                    try:
                        subprocess.run(
                            cmd,
                            shell=True,
                            check=True,
                            stdout=subprocess.DEVNULL,
                            stderr=subprocess.DEVNULL,
                        )
                        # Delete the webm file after successful conversion
                        (self.data_root / webm_video_path).unlink()

                        # Get video metadata using cv2
                        cap = cv2.VideoCapture(output_path)
                        if not cap.isOpened():
                            continue

                        width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
                        height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
                        fps = int(cap.get(cv2.CAP_PROP_FPS))
                        n_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
                        cap.release()

                        caption = item["label"].replace("pretending to ", "")

                        records.append(
                            {
                                "video_path": mp4_video_path,
                                "caption": caption,
                                "height": height,
                                "width": width,
                                "fps": fps,
                                "n_frames": n_frames,
                                "split": split,
                            }
                        )
                        successful_conversions += 1
                    except subprocess.CalledProcessError:
                        print(f"Conversion failed for {webm_video_path}")

                conversion_rate = (successful_conversions / total_videos) * 100
                print(f"Conversion success rate: {conversion_rate:.2f}%")

        # Save as CSV
        metadata_path = self.data_root / self.metadata_path
        metadata_path.parent.mkdir(parents=True, exist_ok=True)
        df = pd.DataFrame.from_records(records)
        df.to_csv(metadata_path, index=False)
        print(f"Created metadata CSV with {len(records)} videos")