up
Browse files
README.md
CHANGED
|
@@ -1,3 +1,168 @@
|
|
| 1 |
---
|
| 2 |
license: afl-3.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: afl-3.0
|
| 3 |
---
|
| 4 |
+
# pdf-layout-chinese: A Chinese document layout PDF dataset
|
| 5 |
+
|
| 6 |
+
### 介绍
|
| 7 |
+
|
| 8 |
+
pdf-layout-chinese是一个中文文档版面分析数据集，面向中文文献类（论文）场景。包含以下10个label：
|
| 9 |
+
|
| 10 |
+
| 正文 | 标题 | 图片 | 图片标题 | 表格 | 表格标题 | 页眉 | 页脚 | 注释 | 公式 |
|
| 11 |
+
| :----: | :-----: | :------: | :--------------: | :-----: | :-------------: | :------: | :------: | :---------: | :--------: |
|
| 12 |
+
| Text | Title | Figure | Figure caption | Table | Table caption | Header | Footer | Reference | Equation |
|
| 13 |
+
|
| 14 |
+
共包含5000张训练集和1000张验证集，分别在train和val目录下。每张图片对应一个同名的标注文件(.json)。
|
| 15 |
+
|
| 16 |
+
样例展示：
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
### 标注格式
|
| 22 |
+
|
| 23 |
+
我们的标注工具是labelme，所以标注格式和labelme格式一致。这里说明一下比较重要的字段。
|
| 24 |
+
|
| 25 |
+
"shapes": shapes字段是一个list，里面有多个dict，每个dict代表一个标注实例。
|
| 26 |
+
|
| 27 |
+
"labels": 类别。
|
| 28 |
+
|
| 29 |
+
"points": 实例标注。因为我们的标注是Polygon形式，所以points里的坐标数量可能大于4。
|
| 30 |
+
|
| 31 |
+
"shape_type": "polygon"
|
| 32 |
+
|
| 33 |
+
"imagePath": 图片路径/名
|
| 34 |
+
|
| 35 |
+
"imageHeight": 高
|
| 36 |
+
|
| 37 |
+
"imageWidth": 宽
|
| 38 |
+
|
| 39 |
+
展示一个完整的标注样例:
|
| 40 |
+
|
| 41 |
+
```
|
| 42 |
+
{
|
| 43 |
+
"version":"4.5.6",
|
| 44 |
+
"flags":{},
|
| 45 |
+
"shapes":[
|
| 46 |
+
{
|
| 47 |
+
"label":"Title",
|
| 48 |
+
"points":[
|
| 49 |
+
[
|
| 50 |
+
553.1111111111111,
|
| 51 |
+
166.59259259259258
|
| 52 |
+
],
|
| 53 |
+
[
|
| 54 |
+
553.1111111111111,
|
| 55 |
+
198.59259259259258
|
| 56 |
+
],
|
| 57 |
+
[
|
| 58 |
+
686.1111111111111,
|
| 59 |
+
198.59259259259258
|
| 60 |
+
],
|
| 61 |
+
[
|
| 62 |
+
686.1111111111111,
|
| 63 |
+
166.59259259259258
|
| 64 |
+
]
|
| 65 |
+
],
|
| 66 |
+
"group_id":null,
|
| 67 |
+
"shape_type":"polygon",
|
| 68 |
+
"flags":{}
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"label":"Text",
|
| 72 |
+
"points":[
|
| 73 |
+
[
|
| 74 |
+
250.5925925925925,
|
| 75 |
+
298.0740740740741
|
| 76 |
+
],
|
| 77 |
+
[
|
| 78 |
+
250.5925925925925,
|
| 79 |
+
345.0740740740741
|
| 80 |
+
],
|
| 81 |
+
[
|
| 82 |
+
188.5925925925925,
|
| 83 |
+
345.0740740740741
|
| 84 |
+
],
|
| 85 |
+
[
|
| 86 |
+
188.5925925925925,
|
| 87 |
+
410.0740740740741
|
| 88 |
+
],
|
| 89 |
+
[
|
| 90 |
+
188.5925925925925,
|
| 91 |
+
456.0740740740741
|
| 92 |
+
],
|
| 93 |
+
[
|
| 94 |
+
324.5925925925925,
|
| 95 |
+
456.0740740740741
|
| 96 |
+
],
|
| 97 |
+
[
|
| 98 |
+
324.5925925925925,
|
| 99 |
+
410.0740740740741
|
| 100 |
+
],
|
| 101 |
+
[
|
| 102 |
+
1051.5925925925926,
|
| 103 |
+
410.0740740740741
|
| 104 |
+
],
|
| 105 |
+
[
|
| 106 |
+
1051.5925925925926,
|
| 107 |
+
345.0740740740741
|
| 108 |
+
],
|
| 109 |
+
[
|
| 110 |
+
1052.5925925925926,
|
| 111 |
+
345.0740740740741
|
| 112 |
+
],
|
| 113 |
+
[
|
| 114 |
+
1052.5925925925926,
|
| 115 |
+
298.0740740740741
|
| 116 |
+
]
|
| 117 |
+
],
|
| 118 |
+
"group_id":null,
|
| 119 |
+
"shape_type":"polygon",
|
| 120 |
+
"flags":{}
|
| 121 |
+
},
|
| 122 |
+
{
|
| 123 |
+
"label":"Footer",
|
| 124 |
+
"points":[
|
| 125 |
+
[
|
| 126 |
+
1033.7407407407406,
|
| 127 |
+
1634.5185185185185
|
| 128 |
+
],
|
| 129 |
+
[
|
| 130 |
+
1033.7407407407406,
|
| 131 |
+
1646.5185185185185
|
| 132 |
+
],
|
| 133 |
+
[
|
| 134 |
+
1052.7407407407406,
|
| 135 |
+
1646.5185185185185
|
| 136 |
+
],
|
| 137 |
+
[
|
| 138 |
+
1052.7407407407406,
|
| 139 |
+
1634.5185185185185
|
| 140 |
+
]
|
| 141 |
+
],
|
| 142 |
+
"group_id":null,
|
| 143 |
+
"shape_type":"polygon",
|
| 144 |
+
"flags":{}
|
| 145 |
+
}
|
| 146 |
+
],
|
| 147 |
+
"imagePath":"val_0031.jpg",
|
| 148 |
+
"imageData":null,
|
| 149 |
+
"imageHeight":1754,
|
| 150 |
+
"imageWidth":1240
|
| 151 |
+
}
|
| 152 |
+
```
|
| 153 |
+
|
| 154 |
+
### 转coco格式
|
| 155 |
+
|
| 156 |
+
执行命令:
|
| 157 |
+
|
| 158 |
+
```
|
| 159 |
+
# train
|
| 160 |
+
python3 labelme2coco.py train train_save_path --labels labels.txt
|
| 161 |
+
|
| 162 |
+
# val
|
| 163 |
+
python3 labelme2coco.py val val_save_path --labels labels.txt
|
| 164 |
+
```
|
| 165 |
+
|
| 166 |
+
转换结果保存在train_save_path/val_save_path目录下。
|
| 167 |
+
|
| 168 |
+
labelme2coco.py取自labelme，更多信息请参考[labelme官方项目](https://github.com/wkentaro/labelme/tree/master/examples/instance_segmentation)
|