README.md

---
dataset_info:
  features:
  - name: image_name
    dtype: string
  - name: image
    dtype: image
  - name: obb
    dtype: string
  splits:
  - name: train
    num_bytes: 446206006
    num_examples: 3053
  - name: validation
    num_bytes: 101618505
    num_examples: 709
  download_size: 547824511
  dataset_size: 547824511
configs:
- config_name: default
  data_files:
  - split: train
    path: data/dataset_with_images_obb_train.parquet
  - split: validation
    path: data/dataset_with_images_obb_val.parquet
license: cc-by-4.0
task_categories:
- object-detection
language:
- km
size_categories:
- 1K<n<10K
---
# Table Dataset - Image & OBB Annotation (Train/Val Split)

## Dataset Overview

Table detection dataset with OBB (Oriented Bounding Box) annotations in YOLO format,
split into training and validation sets.

- **Total examples:** 3762 image-annotation pairs
  - Train: 3053 (81.2%)
  - Validation: 709 (18.8%)
- **Total size:** 522.45 MB
- **Language:** Khmer (km)
- **Document types:** Table documents
- **Annotation format:** YOLO OBB (class cx cy w h)

## Dataset Statistics

### Split Information

| Split       | Examples | Size (MB)     |
|------------|----------|---------------|
| Train      | 3053     | 425.54       |
| Validation | 709      | 96.91        |
| **Total**  | **3762** | **522.45** |

### Train/Val Ratio
- **Train:** 81%
- **Validation:** 19%

## Features

| Feature      | Type          | Description                                      |
|-------------|---------------|--------------------------------------------------|
| `image_name` | string        | Document image filename (without extension)      |
| `image`      | image (bytes) | PNG image binary data                            |
| `obb`        | string        | OBB YOLO annotations (class cx cy w h per line)  |

## Class Names

| ID | Name   |
|----|--------|
| 0  | cell   |
| 1  | column |
| 2  | header |
| 3  | row    |

## Data Format

### Image

PNG binary data — convert to PIL Image for processing:

```python
from PIL import Image
from io import BytesIO

image_bytes = row['image']['bytes']   # HF datasets raw access
image = Image.open(BytesIO(image_bytes))
```

### OBB TXT (string)

YOLO format: one detection per line — `class_id cx cy w h` (all normalised 0-1):

```python
obb_text = row['obb']
for line in obb_text.strip().splitlines():
    parts = line.split()
    class_id = int(parts[0])
    cx, cy, w, h = map(float, parts[1:])
```

## Usage Examples

### Load with pandas

```python
import pandas as pd

df_train = pd.read_parquet('data/dataset_with_images_obb_train.parquet')
df_val   = pd.read_parquet('data/dataset_with_images_obb_val.parquet')
print(f"Train: {len(df_train)}, Val: {len(df_val)}")
```

### Load with Hugging Face Datasets

```python
from datasets import load_dataset

dataset = load_dataset('parquet', data_files={
    'train':      'data/dataset_with_images_obb_train.parquet',
    'validation': 'data/dataset_with_images_obb_val.parquet',
})
```

### Access a single row

```python
from PIL import Image
from io import BytesIO

row = df_train.iloc[0]

image = Image.open(BytesIO(row['image']['bytes']))
print(image.size)          # (width, height)
print(row['image_name'])   # filename stem
print(row['obb'])          # raw YOLO OBB text
```

## File Summary

| File                                    | Rows   | Size (MB)     |
|----------------------------------------|--------|---------------|
| dataset_with_images_obb_train.parquet   | 3053   | 425.54       |
| dataset_with_images_obb_val.parquet     | 709    | 96.91        |

## Citation

```bibtex
@dataset{table_dataset_obb_2026,
  title={Table Dataset - Image & OBB Annotations (Train/Val Split)},
  year={2026},
  note={Table detection dataset with YOLO OBB annotations}
}
```

## License

CC-BY-4.0

---

**Last Updated:** 2026-06-01
**Dataset Version:** 1.0
**Total Examples:** 3762
**Total Size:** 522.45 MB