百度easydl数据标注

一/ 百度easydl数据标注脚本

1 官方标注工具，链接如下，由lableme改进而形成

GitHub - Baidu-AIP/Easyyibiao

2 官网数据导入格式三种分别为：

布局如图所示

2.1json 通用格式 .json

{"labels": [{"y1": 579, "x2": 466, "x1": 328, "y2": 718, "name": "other","meta":{"points":[{"y": 718,"x": 400},{"y": 626,"x": 328},{"y": 579,"x": 393},{"y": 672,"x": 466}]}}]}

2.2 xml ,这个比较好扩展 object节点，复制增加即可

<?xml version="1.0" encoding="utf-8"?>
<annotation><filename>00036.jpg</filename><segmented>0</segmented><owner><name>Lmars, Wuhan University</name><flickrid>I do not know</flickrid></owner><folder>RSDS2016</folder><object><name>other</name><pose>Left</pose><truncated>1</truncated><difficult>0</difficult><quad><x1>400</x1><y1>718</y1><x2>328</x2><y2>626</y2><x3>393</x3><y3>579</y3><x4>466</x4><y4>672</y4></quad><bbox><x1>328</x1><y1>579</y1><x2>466</x2><y2>718</y2></bbox></object>
</annotation>

2.3 coco json

{"info": {"contributor": "nihao", "data_created": "2021", "version": "1.0", "year": 2021}, "licenses": "licenses", "image_nums": 1, "images": [{"file_name": "00036.jpg", "id": 1, "width": 1024, "height": 768}], "categories": [{"id": 1, "name": "other", "supercategory": "other"}], "annotations": [{"category_id": 1, "bbox":[328, 579, 138, 139],"area": 9430, "segmentation": [[400, 718, 328, 626, 393, 579, 466, 672]], "iscrowd": 0, "image_id": 1, "id": 1, "shape": "quad"}]}

3 接下来，通过程序自动生成xml文件

txt2xml.pyimport os
from lxml.etree import Element, SubElement, tostringdef txt_xml(img_name, txt_path, img_xml, xml_path):#读取txt的信息clas=[]imh, imw = 800, 800txt_img=os.path.join(txt_path,img_name)with open(txt_img,"r") as f:for line in f.readlines():line = line.strip('\n')list = line.split(" ")clas.append(list)       # [0, x1, y1, x2, y2]node_root = Element('annotation')node_folder = SubElement(node_root, 'folder')node_folder.text = '1'# filenamenode_filename = SubElement(node_root, 'filename')node_filename.text = img_name.split(".")[0]+".jpg"# pathnode_path = SubElement(node_root, 'path')node_path.text = str(txt_img).split('.')[0] + '.jpg'# sourcenode_source = SubElement(node_root, 'source')node_database = SubElement(node_source, 'database')node_database.text = 'Unknown'# size# node_size = SubElement(node_root, 'size')# node_width = SubElement(node_size, 'width')# node_width.text = str(imw)# node_height = SubElement(node_size, 'height')# node_height.text = str(imh)# node_depth = SubElement(node_size, 'depth')# node_depth.text = '3'# segmentednode_segmented = SubElement(node_root, 'segmented')node_segmented.text = '0'# objectfor i in range(len(clas)):node_object = SubElement(node_root, 'object')node_name = SubElement(node_object, 'name')node_name.text = 'other'node_pose=SubElement(node_object, 'pose')node_pose.text="Left"node_truncated=SubElement(node_object, 'truncated')node_truncated.text="1"node_difficult = SubElement(node_object, 'difficult')node_difficult.text = '0'# bndboxnode_bndbox = SubElement(node_object, 'quad')x1 = SubElement(node_bndbox, 'x1')x1.text = str(clas[i][1])y1 = SubElement(node_bndbox, 'y1')y1.text = str(clas[i][2])x2 = SubElement(node_bndbox, 'x2')x2.text = str(clas[i][3])y2 = SubElement(node_bndbox, 'y2')y2.text = str(clas[i][4])x3 = SubElement(node_bndbox, 'x3')x3.text = str(clas[i][5])y3 = SubElement(node_bndbox, 'y3')y3.text = str(clas[i][6])x4 = SubElement(node_bndbox, 'x4')x4.text = str(clas[i][7])y4 = SubElement(node_bndbox, 'y4')y4.text = str(clas[i][8])  xml = tostring(node_root, pretty_print=True)  # 格式化显示，该换行的换行img_newxml = os.path.join(xml_path, img_xml)file_object = open(img_newxml, 'wb')file_object.write(xml)file_object.close()if __name__ == "__main__":#标注文件夹所在位置txt_path=r"temp"#txt转化成xml格式后存放的文件夹xml_path=r"temp1"if not os.path.exists(xml_path):os.mkdir(xml_path)for img_name in os.listdir(txt_path):print(img_name)img_xml=img_name.split(".")[0]+".xml"txt_xml(img_name, txt_path, img_xml, xml_path)

最终效果图

参考 yolo图像检测数据集格式转换：xml 与 txt格式相互转换_uncle_ll的博客-CSDN博客_yolo数据集txt格式

二/ 利用PIL 模块生成相应的图片字符图片

参考link ：https://github.com/mpcabd/python-arabic-reshaper

https://github.com/MichalBusta/E2E-MLT

效果：

pip install --upgrade arabic-reshaper

conda install -c mpcabd arabic-reshaper

pip install --upgrade arabic-reshaper python-bidi pillow

代码

#产生阿拉伯文图片
import arabic_reshapertext_to_be_reshaped = '2023 06 14/2022 06 16'
text_to_be_reshaped1='JXG'
text_to_be_reshaped2='14 06 2023/16 06 2022 X21'reshaped_text = arabic_reshaper.reshape(text_to_be_reshaped)'''
At this stage the text is reshaped, all letters are in their correct form
based on their surroundings, but if you are going to print the text in a
left-to-right context, which usually happens in libraries/apps that do not
support Arabic and/or right-to-left text rendering, then you need to use
get_display from python-bidi.
Note that this is optional and depends on your usage of the reshaped text.
'''
from bidi.algorithm import get_display
bidi_text = get_display(reshaped_text)# At this stage the text in bidi_text can be easily rendered in any library
# that doesn't support Arabic and/or right-to-left, so use it as you'd use
# any other string. For example if you're using PIL.ImageDraw.text to draw
# text over an image you'd just use it like this...from PIL import Image, ImageDraw, ImageFont# We load Arial since it's a well known font that supports Arabic Unicode
# font = ImageFont.truetype('Arial', 40)
font = ImageFont.truetype('/PaddleOCR/StyleText/fonts/arabic.ttf', 50)
font1= ImageFont.truetype('PaddleOCR/StyleText/fonts/en_standard.ttf',40)
image = Image.new('RGBA', (800, 600), (255,255,255,0))
image_draw = ImageDraw.Draw(image)
image_draw.text((350,10), text_to_be_reshaped1, fill=(255,255,255,200), font=font1)
image_draw.text((10,10), bidi_text, fill=(255,255,255,200), font=font)
image_draw.text((10,70), text_to_be_reshaped2, fill=(255,255,255,200), font=font1)# image.show()
image.save("temp.png")

百度easydl数据标注相关推荐

《新闻联播》报道百度山西数据标注基地培育新业态打开新就业空间
人工智能通过培育新业态创造出更多的新就业方式,正在为中国稳就业作出重要贡献. 7月26日,央视<新闻联播>--"培育新业态打开就业新空间"报道,全国城镇调查失业率实现 ...
基于百度EasyDL定制化图像识别平台的海洋鱼类识别方法
[目的]鱼类识别对渔业资源的开发利用有着重要的意义.针对海底环境恶劣.拍摄环境亮度低.场景模糊的实际情况导致海底观测视频品质差,视频中的鱼类识别难的问题以及现有鱼类识别方法存在的鱼类标注数据集过少导致 ...
@赛迪顾问拍了拍我：数据标注基地全国多点开花
近日,国内权威咨询机构赛迪顾问发布<百度智能云人工智能基础数据产业基地项目价值评估报告>(以下简称报告).报告显示,作为 AI 新业态,发展人工智能基础数据服务产业对培育人工智能产业,推动 ...
基于JDBC从数据库中读取数据，在百度地图批量标注地点
基于JDBC从数据库中读取数据,在百度地图批量标注地点一.相关技术 JSP,JDBC,JSON,JS,百度地图API 二.基于JDBC从数据库读取数据见"JSP基于JDBC操作MSSQL ...
从谷歌AutoML到百度EasyDL，AI大生产时代，调参师不再是刚需
出品 | AI科技大本营头图 | 付费下载于视觉中国 2018 年,Google Cloud 宣布将 AutoML 作为机器学习产品的一部分.至此,AutoML 开始进入大众的视野. 实际上,201 ...
如何获取高精度CV模型？快来试试百度EasyDL超大规模视觉预训练模型
在深度学习领域,有一个名词正在被越来越频繁地得到关注:迁移学习.它相比效果表现好的监督学习来说,可以减去大量的枯燥标注过程,简单来说就是在大数据集训练的预训练模型上进行小数据集的迁移,以获得对新数据较 ...
Paw 百度ai_直面落地！百度EasyDL产业智能创新大赛成果覆盖能源、交通、水利民生重业...
物体检测模型实现高压线路隐患检测.图像分类实现短视频快速剪辑和量产.文本情感分类辅助潜在心理疾病患者自发检测.图片识别车辆轮轴数监管车辆载重--每一个创想都能简单快速实现,没有AI开发基础的小伙伴们也 ...
百度EasyDL深度学习实战营，免费教你转型AI工程师！
新型冠状病毒的疫情动态牵动着每个人的心,在当下疫情控制阶段,AI技术的应用也正在为抗疫贡献自己的力量. 百度飞桨联合百度技术学院正式加推"EasyDL深度学习实战营"系列直播课,免 ...
如何通过大华sdk采集一帧图像?_EasyData解放数据标注员双手，采集清洗标注一站搞定...
在AI模型开发中,数据准备往往占据了大量时间.经过长时间的调研与访问,我们发现用户常常会遇到以下问题: · 难以获取与场景数据匹配的训练数据.在具体业务场景中,模型的效果至关重要,我们往往会追求高精度 ...

百度easydl数据标注

百度easydl数据标注相关推荐

最新文章

热门文章