---
license: mit
task_categories:
- object-detection
language:
- zh
- en
pretty_name: LottoScan License Plates
tags:
- license-plate
- ocr
- obb
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: D_01
    path: data/D_01/*.parquet
  - split: D_05
    path: data/D_05/*.parquet
---

# LottoScan License Plates

License plate dataset with **4-point oriented bounding boxes** and plate text labels.
All images are 640x640 JPEG.

Total samples: **2**
Splits: **24** (D-01..D-12 for daytime, N-01..N-12 for nighttime)

## Categories (4 dims × cartesian)

Split names use underscore (D_NN, N_NN) because HuggingFace forbids
dashes in split names.

- Time: D=day / N=night
- Distance: near / mid / far (encoded in id index 1-4 / 5-8 / 9-12)
- Plate count: 1-2 (odd within block) / 3+ (even within block)
- Angle: front <30° (odd) / oblique >30° (even)

Per-split row counts:

| Split | Rows |
|-------|------|
| D_01 | 1 |
| D_02 | 0 |
| D_03 | 0 |
| D_04 | 0 |
| D_05 | 1 |
| D_06 | 0 |
| D_07 | 0 |
| D_08 | 0 |
| D_09 | 0 |
| D_10 | 0 |
| D_11 | 0 |
| D_12 | 0 |
| N_01 | 0 |
| N_02 | 0 |
| N_03 | 0 |
| N_04 | 0 |
| N_05 | 0 |
| N_06 | 0 |
| N_07 | 0 |
| N_08 | 0 |
| N_09 | 0 |
| N_10 | 0 |
| N_11 | 0 |
| N_12 | 0 |

## Schema (per row)

| Field            | Type           | Description                                      |
|------------------|----------------|--------------------------------------------------|
| `image`          | Image          | 640x640 JPEG                                     |
| `quadrilaterals` | float[N][4][2] | Normalized 4-point quads, order: 左上→右上→右下→左下 |
| `plate_texts`    | string[N]      | Plate text strings, parallel to quadrilaterals   |

## Usage

```python
from datasets import load_dataset
ds_d05 = load_dataset("a27259954/4DataTool", split="D_05")  # 中距離 / 1-2 個 / 直視
sample = ds_d05[0]
sample["image"]            # PIL.Image
sample["quadrilaterals"]   # [[[x,y],...]]
sample["plate_texts"]      # ["ABC-1234", ...]
```