File size: 4,327 Bytes
50fd1e7
d08bdd9
50fd1e7
d08bdd9
50fd1e7
 
 
d08bdd9
50fd1e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d08bdd9
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
---
license: apache-2.0
task_categories:
- feature-extraction
language:
- ja
size_categories:
- n<1K
---

# Japanese TV Commercial Video Dataset

## Dataset Description

This dataset contains Japanese TV commercial (CM) videos with hierarchical scene detection annotations.

### Dataset Summary

- **Total Videos**: 100
- **Total Duration**: 39.5 minutes (2370 seconds)
- **Total Scenes**: 2,320
- **Average Duration per Video**: 23.7s
- **Average Scenes per Video**: 23.2

### Languages

Commercial videos contain a mix of:
- Japanese (primary)
- English (secondary)

## Dataset Structure

```
dataset/
├── videos/              # Video files (.mp4)
│   ├── CM_000.mp4
│   ├── CM_001.mp4
│   └── ...
├── thumbnails/          # Representative thumbnails for each video
│   ├── CM_000.jpg
│   ├── CM_001.jpg
│   └── ...
├── scenes/              # Scene detection results
│   ├── CM_000/
│   │   ├── scenes.json  # Scene boundaries and metadata
│   │   └── scene_*.jpg  # Thumbnails for each scene
│   └── ...
├── metadata.parquet     # Dataset metadata (recommended)
└── metadata.csv         # Human-readable metadata

```

### Data Fields

**metadata.parquet / metadata.csv**:
- `video_id`: Unique identifier (e.g., "CM_000")
- `video_path`: Path to video file
- `thumbnail`: Path to representative thumbnail image
- `duration`: Video duration in seconds
- `num_scenes`: Total number of detected scenes
- `num_groups`: Number of scene groups
- `scenes_json_path`: Path to scenes.json file
- `scene_thumbnails_dir`: Directory containing scene thumbnails
- `level1_count`, `level2_count`, `level3_count`: Scene counts per detection level
- `level1_threshold`, `level2_threshold`, `level3_threshold`: Detection thresholds used

**scenes.json** (per video):
```json
{
  "video_name": "CM_000",
  "video_duration": 29.66,
  "total_scenes": 15,
  "scenes": [
    {
      "scene_number": 1,
      "start_time": 0.0,
      "end_time": 2.102,
      "duration": 2.102,
      "start_timecode": "00:00:00.000",
      "end_timecode": "00:00:02.102",
      "thumbnail": "scene_001.jpg",
      "level": 1,
      "threshold": 5.0
    }
  ]
}
```

## Scene Detection Methodology

Scenes are detected using hierarchical scene detection with PySceneDetect:
- **Level 1** (threshold: 5.0): Major scene changes (coarse)
- **Level 2** (threshold: 3.0): Medium scene changes
- **Level 3** (threshold: 1.0): Subtle scene changes (fine)

Each scene includes:
- Precise start/end timestamps
- Duration
- Detection level (indicating cut intensity)
- Thumbnail image

## Usage

### Load with Hugging Face Datasets

```python
from datasets import load_dataset

# Load metadata
dataset = load_dataset("your-username/tv-commercial-videos")

# Access first video
sample = dataset["train"][0]
print(f"Video: {sample['video_path']}")
print(f"Duration: {sample['duration']}s")
print(f"Scenes: {sample['num_scenes']}")
```

### Load with Pandas

```python
import pandas as pd

# Load metadata
df = pd.read_parquet("metadata.parquet")

# Load scene data for specific video
import json
with open(df.iloc[0]['scenes_json_path'], 'r') as f:
    scenes = json.load(f)
```

## Use Cases

This dataset is suitable for:
- **Video segmentation**: Scene boundary detection
- **Content analysis**: Commercial structure analysis
- **Computer vision**: Object detection in commercial contexts
- **Temporal analysis**: Shot duration patterns
- **Multi-modal learning**: Video + audio + text
- **Advertisement research**: Creative patterns in commercials

## Limitations

- Videos are sourced from Japanese TV commercials (specific domain)
- Scene detection is automated and may have occasional errors
- No manual verification of scene boundaries
- No semantic labels (e.g., product categories, themes)

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{tv_commercial_videos_2024,
  title={Japanese TV Commercial Video Dataset with Scene Detection},
  author={Your Name},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/your-username/tv-commercial-videos}
}
```

## License

This dataset is released under CC-BY-4.0 license.

## Contact

For questions or issues, please open an issue on the dataset repository.