PaddleOCRv3之二：TextRecognitionDataGenerator训练集构造

OCR识别部分数据集中字体和背景是比较重要的方面，在实际的场景下收集不到那么多真实的样本，在训练开始的时候手工构造一批训练数据还是很有必要的。可以用这一批数据预训练后面用真实数据来微调，也可以直接把这些数据和真实数据混合在一起训练。
在构造数据集的方面TextRecognitionDataGenerator还是比较好用了，这是按行生成文本的,可以生成bounding boxes，和mask。

1. 开源项目TextRecognitionDataGenerator

链接TextRecognitionDataGenerator

支持的一些有用的操作

• blur 模糊操作
• 在背景图上写字（这个很有用，选择一些真实场景的背景图，然后把文字写在上面，看起来更真实一些）
• 角度旋转
• 生成boundingbox和mask标签
• 字间距调整
• 字体颜色设置
• 拉丁语系的提供了100多种字体
• …

安装有两种方式，

• pip 安装

pip install trdg
1

• 源码安装
  下载源码，cd进入setup.py所在目录，然后安装依赖项

pip install -r requirements.txt
1

使用的时候用trdg文件夹下的run.py文件生成数据，默认在TextRecognitionDataGenerator-master\trdg路径下

2. 使用

常用选项

python run.py 
--font_dir fonts\latin 	#字体文件，可以选择文件夹或者单个字体库
--dict dicts\myDict.txt 	#字典路径
-c 50 		#一共生成多少个图片
--output_dir outputs #保存路径
-k 5 -rk  		#-k 5表示旋转的角度为5°，后面接-rk表示在-5，+5范围内随机，
-bl 3 -rbl		#-bl 表示blur 高斯模糊，后面接半径， -rbl表示高斯核的半径在0-3之间
--case upper 	#upper表示使用大写字符，lower表示用小写
-b 3 -id images	#-b表示背景，3表示用图片做背景，-id：image_dir,指定背景图片路径
-f 64		# --format ,生成的图片的高度
-tc #22211f	#--text_color 6位的16进制数，RGB格式直接翻译，例如：rgb=[10,15,255]==>#0a 0e ff 
-obb 1		#生成bonding box，
		#格式是嵌套的，第一行：4个数，分别为第一个字符左上角x,y 最后一个字符的右下角的x，y
		#第二行：4个数，分别为第二个字符左上角的x,y，最后一个字符的右下角的x，y
		#第n行：4个数，分别为第n个字符左上角的x,y，最后一个字符的右下角的x，y

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

例如：

python run.py --font_dir fonts\latin --dict dicts\OCRDict.txt -c 52 --output_dir K:\imageData\OCR\ocr_dataset\test  -k 6 -rk -bl 1 -rbl -b 3 -id K:\imageData\OCR\ocr_background\yellow -f 80
1

查看使用帮助就能只知道完整的用法了

python run.py --help
1

usage: run.py [-h] [--output_dir [OUTPUT_DIR]] [-i [INPUT_FILE]]
              [-l [LANGUAGE]] -c [COUNT] [-rs] [-let] [-num] [-sym]
              [-w [LENGTH]] [-r] [-f [FORMAT]] [-t [THREAD_COUNT]]
              [-e [EXTENSION]] [-k [SKEW_ANGLE]] [-rk] [-wk] [-bl [BLUR]]
              [-rbl] [-b [BACKGROUND]] [-hw] [-na NAME_FORMAT]
              [-om OUTPUT_MASK] [-obb OUTPUT_BBOXES] [-d [DISTORSION]]
              [-do [DISTORSION_ORIENTATION]] [-wd [WIDTH]] [-al [ALIGNMENT]]
              [-or [ORIENTATION]] [-tc [TEXT_COLOR]] [-sw [SPACE_WIDTH]]
              [-cs [CHARACTER_SPACING]] [-m [MARGINS]] [-fi] [-ft [FONT]]
              [-fd [FONT_DIR]] [-id [IMAGE_DIR]] [-ca [CASE]] [-dt [DICT]]
              [-ws] [-stw [STROKE_WIDTH]] [-stf [STROKE_FILL]]
              [-im [IMAGE_MODE]]

Generate synthetic text data for text recognition.

optional arguments:
  -h, --help            show this help message and exit
  --output_dir [OUTPUT_DIR]
                        The output directory
  -i [INPUT_FILE], --input_file [INPUT_FILE]
                        When set, this argument uses a specified text file as
                        source for the text
  -l [LANGUAGE], --language [LANGUAGE]
                        The language to use, should be fr (French), en
                        (English), es (Spanish), de (German), ar (Arabic), cn
                        (Chinese), ja (Japanese) or hi (Hindi)
  -c [COUNT], --count [COUNT]
                        The number of images to be created.
  -rs, --random_sequences
                        Use random sequences as the source text for the
                        generation. Set '-let','-num','-sym' to use
                        letters/numbers/symbols. If none specified, using all
                        three.
  -let, --include_letters
                        Define if random sequences should contain letters.
                        Only works with -rs
  -num, --include_numbers
                        Define if random sequences should contain numbers.
                        Only works with -rs
  -sym, --include_symbols
                        Define if random sequences should contain symbols.
                        Only works with -rs
  -w [LENGTH], --length [LENGTH]
                        Define how many words should be included in each
                        generated sample. If the text source is Wikipedia,
                        this is the MINIMUM length
  -r, --random          Define if the produced string will have variable word
                        count (with --length being the maximum)
  -f [FORMAT], --format [FORMAT]
                        Define the height of the produced images if
                        horizontal, else the width
  -t [THREAD_COUNT], --thread_count [THREAD_COUNT]
                        Define the number of thread to use for image
                        generation
  -e [EXTENSION], --extension [EXTENSION]
                        Define the extension to save the image with
  -k [SKEW_ANGLE], --skew_angle [SKEW_ANGLE]
                        Define skewing angle of the generated text. In
                        positive degrees
  -rk, --random_skew    When set, the skew angle will be randomized between
                        the value set with -k and it's opposite
  -wk, --use_wikipedia  Use Wikipedia as the source text for the generation,
                        using this paremeter ignores -r, -n, -s
  -bl [BLUR], --blur [BLUR]
                        Apply gaussian blur to the resulting sample. Should be
                        an integer defining the blur radius
  -rbl, --random_blur   When set, the blur radius will be randomized between 0
                        and -bl.
  -b [BACKGROUND], --background [BACKGROUND]
                        Define what kind of background to use. 0: Gaussian
                        Noise, 1: Plain white, 2: Quasicrystal, 3: Image
  -hw, --handwritten    Define if the data will be "handwritten" by an RNN
  -na NAME_FORMAT, --name_format NAME_FORMAT
                        Define how the produced files will be named. 0:
                        [TEXT]_[ID].[EXT], 1: [ID]_[TEXT].[EXT] 2: [ID].[EXT]
                        + one file labels.txt containing id-to-label mappings
  -om OUTPUT_MASK, --output_mask OUTPUT_MASK
                        Define if the generator will return masks for the text
  -obb OUTPUT_BBOXES, --output_bboxes OUTPUT_BBOXES
                        Define if the generator will return bounding boxes for
                        the text, 1: Bounding box file, 2: Tesseract format
  -d [DISTORSION], --distorsion [DISTORSION]
                        Define a distorsion applied to the resulting image. 0:
                        None (Default), 1: Sine wave, 2: Cosine wave, 3:
                        Random
  -do [DISTORSION_ORIENTATION], --distorsion_orientation [DISTORSION_ORIENTATION]
                        Define the distorsion's orientation. Only used if -d
                        is specified. 0: Vertical (Up and down), 1: Horizontal
                        (Left and Right), 2: Both
  -wd [WIDTH], --width [WIDTH]
                        Define the width of the resulting image. If not set it
                        will be the width of the text + 10. If the width of
                        the generated text is bigger that number will be used
  -al [ALIGNMENT], --alignment [ALIGNMENT]
                        Define the alignment of the text in the image. Only
                        used if the width parameter is set. 0: left, 1:
                        center, 2: right
  -or [ORIENTATION], --orientation [ORIENTATION]
                        Define the orientation of the text. 0: Horizontal, 1:
                        Vertical
  -tc [TEXT_COLOR], --text_color [TEXT_COLOR]
                        Define the text's color, should be either a single hex
                        color or a range in the ?,? format.
  -sw [SPACE_WIDTH], --space_width [SPACE_WIDTH]
                        Define the width of the spaces between words. 2.0
                        means twice the normal space width
  -cs [CHARACTER_SPACING], --character_spacing [CHARACTER_SPACING]
                        Define the width of the spaces between characters. 2
                        means two pixels
  -m [MARGINS], --margins [MARGINS]
                        Define the margins around the text when rendered. In
                        pixels
  -fi, --fit            Apply a tight crop around the rendered text
  -ft [FONT], --font [FONT]
                        Define font to be used
  -fd [FONT_DIR], --font_dir [FONT_DIR]
                        Define a font directory to be used
  -id [IMAGE_DIR], --image_dir [IMAGE_DIR]
                        Define an image directory to use when background is
                        set to image
  -ca [CASE], --case [CASE]
                        Generate upper or lowercase only. arguments: upper or
                        lower. Example: --case upper
  -dt [DICT], --dict [DICT]
                        Define the dictionary to be used
  -ws, --word_split     Split on words instead of on characters (preserves
                        ligatures, no character spacing)
  -stw [STROKE_WIDTH], --stroke_width [STROKE_WIDTH]
                        Define the width of the strokes
  -stf [STROKE_FILL], --stroke_fill [STROKE_FILL]
                        Define the color of the contour of the strokes, if
                        stroke_width is bigger than 0
  -im [IMAGE_MODE], --image_mode [IMAGE_MODE]
                        Define the image mode to be used. RGB is default, L
                        means 8-bit grayscale images, 1 means 1-bit binary
                        images stored with one pixel per byte, etc.
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

3. 示例

• 下面这几张图是原项目中的示例图

• 下面的图片使用的是真实场景下的一些背景图，按字典把字符写在上面