README.md 6.6 KB
Newer Older
G
grasswolfs 已提交
1
English | [简体中文](README_ch.md)
W
WenmuZhou 已提交
2

W
WenmuZhou 已提交
3

G
grasswolfs 已提交
4
# Getting Started
W
WenmuZhou 已提交
5

M
MissPenguin 已提交
6
[1. Install whl package](#Install)
W
WenmuZhou 已提交
7

M
MissPenguin 已提交
8
[2. Quick Start](#QuickStart)
W
WenmuZhou 已提交
9

G
grasswolfs 已提交
10
[3. PostProcess](#PostProcess)
W
WenmuZhou 已提交
11

G
grasswolfs 已提交
12
[4. Results](#Results)
W
WenmuZhou 已提交
13

G
grasswolfs 已提交
14 15
[5. Training](#Training)

M
MissPenguin 已提交
16
<a name="Install"></a>
G
grasswolfs 已提交
17 18

## 1.  Install whl package
W
WenmuZhou 已提交
19
```bash
G
grasswolfs 已提交
20 21
wget https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl
pip install -U layoutparser-0.0.0-py3-none-any.whl
W
WenmuZhou 已提交
22 23
```

M
MissPenguin 已提交
24
<a name="QuickStart"></a>
W
WenmuZhou 已提交
25

G
grasswolfs 已提交
26
## 2. Quick Start
W
WenmuZhou 已提交
27

G
grasswolfs 已提交
28
Use LayoutParser to identify the layout of a given document:
W
WenmuZhou 已提交
29 30

```python
W
WenmuZhou 已提交
31
import cv2
W
WenmuZhou 已提交
32
import layoutparser as lp
W
WenmuZhou 已提交
33
image = cv2.imread("doc/table/layout.jpg")
W
WenmuZhou 已提交
34 35
image = image[..., ::-1]

G
grasswolfs 已提交
36 37
# load model
model = lp.PaddleDetectionLayoutModel(config_path="lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config",
W
WenmuZhou 已提交
38 39
                                threshold=0.5,
                                label_map={0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"},
G
grasswolfs 已提交
40
                                enforce_cpu=False,
W
WenmuZhou 已提交
41
                                enable_mkldnn=True)
G
grasswolfs 已提交
42
# detect
W
WenmuZhou 已提交
43 44
layout = model.detect(image)

G
grasswolfs 已提交
45
# show result
W
WenmuZhou 已提交
46 47
show_img = lp.draw_box(image, layout, box_width=3, show_element_type=True)
show_img.show()
W
WenmuZhou 已提交
48 49
```

G
grasswolfs 已提交
50
The following figure shows the result, with different colored detection boxes representing different categories and displaying specific categories in the upper left corner of the box with `show_element_type`
W
WenmuZhou 已提交
51 52 53 54

<div align="center">
<img src="../../doc/table/result_all.jpg"  width = "600" />
</div>
G
grasswolfs 已提交
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69
`PaddleDetectionLayoutModel`parameters are described as follows:

|   parameter    |                       description                        |   default   |                            remark                            |
| :------------: | :------------------------------------------------------: | :---------: | :----------------------------------------------------------: |
|  config_path   |                    model config path                     |    None     | Specify config_ path will automatically download the model (only for the first time,the model will exist and will not be downloaded again) |
|   model_path   |                        model path                        |    None     | local model path, config_ path and model_ path must be set to one, cannot be none at the same time |
|   threshold    |              threshold of prediction score               |     0.5     |                              \                               |
|  input_shape   |                 picture size of reshape                  | [3,640,640] |                              \                               |
|   batch_size   |                    testing batch size                    |      1      |                              \                               |
|   label_map    |                  category mapping table                  |    None     | Setting config_ path, it can be none, and the label is automatically obtained according to the dataset name_ map |
|  enforce_cpu   |                    whether to use CPU                    |    False    |      False to use GPU, and True to force the use of CPU      |
| enforce_mkldnn | whether mkldnn acceleration is enabled in CPU prediction |    True     |                              \                               |
|   thread_num   |                the number of CPU threads                 |     10      |                              \                               |

The following model configurations and label maps are currently supported, which you can use by modifying '--config_path' and '--label_map' to detect different types of content:
W
WenmuZhou 已提交
70 71 72 73 74 75 76

| dataset                                                      | config_path                                                  | label_map                                                 |
| ------------------------------------------------------------ | ------------------------------------------------------------ | --------------------------------------------------------- |
| [TableBank](https://doc-analysis.github.io/tablebank-page/index.html) word | lp://TableBank/ppyolov2_r50vd_dcn_365e_tableBank_word/config | {0:"Table"}                                               |
| TableBank latex                                              | lp://TableBank/ppyolov2_r50vd_dcn_365e_tableBank_latex/config | {0:"Table"}                                               |
| [PubLayNet](https://github.com/ibm-aur-nlp/PubLayNet)        | lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config      | {0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"} |

G
grasswolfs 已提交
77 78
* TableBank word and TableBank latex are trained on datasets of word documents and latex documents respectively;
* Download TableBank dataset contains both word and latex。
W
WenmuZhou 已提交
79

G
grasswolfs 已提交
80
<a name="PostProcess"></a>
W
WenmuZhou 已提交
81

G
grasswolfs 已提交
82
## 3. PostProcess
W
WenmuZhou 已提交
83

G
grasswolfs 已提交
84
Layout parser contains multiple categories, if you only want to get the detection box for a specific category (such as the "Text" category), you can use the following code:
W
WenmuZhou 已提交
85 86

```python
G
grasswolfs 已提交
87 88
# follow the above code
# filter areas for a specific text type
W
WenmuZhou 已提交
89 90 91
text_blocks = lp.Layout([b for b in layout if b.type=='Text'])
figure_blocks = lp.Layout([b for b in layout if b.type=='Figure'])

G
grasswolfs 已提交
92
# text areas may be detected within the image area, delete these areas
W
WenmuZhou 已提交
93 94 95
text_blocks = lp.Layout([b for b in text_blocks \
                   if not any(b.is_in(b_fig) for b_fig in figure_blocks)])

G
grasswolfs 已提交
96
# sort text areas and assign ID
W
WenmuZhou 已提交
97 98 99 100 101 102 103 104 105 106
h, w = image.shape[:2]

left_interval = lp.Interval(0, w/2*1.05, axis='x').put_on_canvas(image)

left_blocks = text_blocks.filter_by(left_interval, center=True)
left_blocks.sort(key = lambda b:b.coordinates[1])

right_blocks = [b for b in text_blocks if b not in left_blocks]
right_blocks.sort(key = lambda b:b.coordinates[1])

G
grasswolfs 已提交
107
# the two lists are merged and the indexes are added in order
W
WenmuZhou 已提交
108 109
text_blocks = lp.Layout([b.set(id = idx) for idx, b in enumerate(left_blocks + right_blocks)])

G
grasswolfs 已提交
110
# display result
W
WenmuZhou 已提交
111
show_img = lp.draw_box(image, text_blocks,
G
grasswolfs 已提交
112
            box_width=3,
W
WenmuZhou 已提交
113
            show_element_id=True)
W
WenmuZhou 已提交
114
show_img.show()
W
WenmuZhou 已提交
115 116
```

G
grasswolfs 已提交
117
Displays results with only the "Text" category:
W
WenmuZhou 已提交
118 119 120 121

<div align="center">
<img src="../../doc/table/result_text.jpg"  width = "600" />
</div>
G
grasswolfs 已提交
122
<a name="Results"></a>
W
WenmuZhou 已提交
123

G
grasswolfs 已提交
124
## 4. Results
W
WenmuZhou 已提交
125 126 127 128 129 130

| Dataset   | mAP  | CPU time cost | GPU time cost |
| --------- | ---- | ------------- | ------------- |
| PubLayNet | 93.6 | 1713.7ms      | 66.6ms        |
| TableBank | 96.2 | 1968.4ms      | 65.1ms        |

G
grasswolfs 已提交
131
**Envrionment:**
W
WenmuZhou 已提交
132

G
grasswolfs 已提交
133
**CPU:**  Intel(R) Xeon(R) CPU E5-2650 v4 @ 2.20GHz,24core
W
WenmuZhou 已提交
134

G
grasswolfs 已提交
135
**GPU:**  a single NVIDIA Tesla P40
W
WenmuZhou 已提交
136

G
grasswolfs 已提交
137
<a name="Training"></a>
W
WenmuZhou 已提交
138

G
grasswolfs 已提交
139
## 5. Training
W
WenmuZhou 已提交
140

G
grasswolfs 已提交
141
The above model is based on PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection) ,if you want to train your own layout parser model,please refer to:[train_layoutparser_model](train_layoutparser_model.md)