File size: 3,617 Bytes
54a8364
414548b
 
 
e4c0104
414548b
 
 
 
 
 
 
 
 
 
 
 
 
54a8364
414548b
e4c0104
414548b
e4c0104
414548b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e4c0104
414548b
 
 
e4c0104
 
 
414548b
e4c0104
414548b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: cc-by-4.0
language:
  - km
  - en
task_categories:
  - object-detection
tags:
  - khmer
  - text-detection
  - ocr
  - document-analysis
  - yolo
  - synthetic
  - scene-text
pretty_name: Khmer Text Detection (Ultimate)
size_categories:
  - 10K<n<100K
---

# Textline Detection — Ultimate Dataset

A large-scale, multi-source dataset for **textline detection** using YOLO-format bounding box annotations.
Combines real scene text, document layout images, and synthetically generated Khmer document images.

---

## Dataset Summary

| Split | Images |
|-------|--------|
| Train | 30,658 |
| Val   | 2,764 |
| **Total** | **33,422** |

### Classes

| ID | Name | Description |
|----|------|-------------|
| 0 | `text_line` | A line of Khmer or mixed-script text |
| 1 | `image` | An embedded image/figure region within a document |

---

## Data Sources

### 1. Real Scene Text
Images captured in natural environments — street signs, storefronts, billboards, and handwritten Khmer documents.
Derived from the **DonkeySmall** base dataset (~26K images).

### 2. Document Layout
Images from the **DocLayNet** multi-class document layout corpus, re-labelled for the two-class schema.
Covers documents: official reports, newspapers, and books.

### 3. Synthetic Khmer Documents
Programmatically generated document images using custom Khmer text rendering pipelines.
Fonts, sizes, backgrounds, and layouts are randomised. Labels are auto-generated (zero annotation cost).

---

## Annotation Format

Labels follow **YOLO v8** format — coordinates normalised to `[0, 1]`:

```
<class_id> <cx> <cy> <width> <height>
```

The `annotations` field is a JSON-serialised list:

```json
[
  {"class_id": 0, "cx": 0.512, "cy": 0.234, "w": 0.310, "h": 0.045},
  {"class_id": 1, "cx": 0.720, "cy": 0.600, "w": 0.200, "h": 0.250}
]
```

---

## Dataset Fields

| Field | Type | Description |
|-------|------|-------------|
| `image` | `Image` | Decoded PIL image |
| `image_path` | `string` | Original file path at collection time |
| `source` | `string` | `real_scene_text` / `doclaynet_khmer` / `synthetic_khmer_doc` |
| `split` | `string` | `train` or `val` |
| `annotations` | `string` | JSON list of YOLO bounding boxes |
| `num_objects` | `int32` | Number of annotated objects |

---

## Usage

```python
from datasets import load_dataset

ds = load_dataset("Darayut/Multilingual-Textline-Detection-Dataset")

sample = ds["train"][0]
print(sample["source"], sample["num_objects"])
sample["image"].show()
```

### Convert back to YOLO label files

```python
import json
from pathlib import Path

def save_labels(split, out_dir):
    Path(out_dir).mkdir(parents=True, exist_ok=True)
    for row in ds[split]:
        stem = Path(row["image_path"]).stem
        anns = json.loads(row["annotations"])
        with open(f"{out_dir}/{stem}.txt", "w") as f:
            for a in anns:
                f.write(f"{a['class_id']} {a['cx']:.6f} {a['cy']:.6f} {a['w']:.6f} {a['h']:.6f}\n")

save_labels("train", "labels/train")
save_labels("val",   "labels/val")
```

### YAML config for YOLOv8 training

```yaml
train: /path/to/images/train
val:   /path/to/images/val
nc: 2
names:
  0: text_line
  1: image
```

---

## Limitations

- Synthetic images may not capture all real-world degradation (blur, skew, lighting variation).
- Scene-text labels are semi-automatic and may have occasional missed detections.
- Dataset is primarily Khmer script; other scripts appear only incidentally.

---

## License

[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) — free to use, share, and adapt with attribution.

---