{"id":13543234,"url":"https://github.com/huoyijie/AdvancedEAST","last_synced_at":"2025-04-02T12:31:48.828Z","repository":{"id":151681191,"uuid":"133290200","full_name":"huoyijie/AdvancedEAST","owner":"huoyijie","description":" AdvancedEAST is an algorithm used for Scene image text detect, which is primarily based on EAST, and the significant improvement was also made, which make long text predictions more accurate.https://github.com/huoyijie/raspberrypi-car","archived":false,"fork":false,"pushed_at":"2022-09-09T02:02:32.000Z","size":3293,"stargazers_count":1227,"open_issues_count":80,"forks_count":381,"subscribers_count":42,"default_branch":"master","last_synced_at":"2025-03-25T14:05:31.256Z","etag":null,"topics":["advancedeast","advancedeast-network-arch","algorithm","bellow","computer-vision","deep-learning","east","icpr","keras","machine-learning","python","scene","tensorflow","text-detect","text-predictions","tian-chi","tianchi"],"latest_commit_sha":null,"homepage":"https://huoyijie.cn/","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/huoyijie.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null}},"created_at":"2018-05-14T01:37:52.000Z","updated_at":"2025-01-19T03:40:30.000Z","dependencies_parsed_at":null,"dependency_job_id":"d61aed20-9b6e-45b6-99b4-5b9b15213772","html_url":"https://github.com/huoyijie/AdvancedEAST","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/huoyijie%2FAdvancedEAST","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/huoyijie%2FAdvancedEAST/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/huoyijie%2FAdvancedEAST/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/huoyijie%2FAdvancedEAST/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/huoyijie","download_url":"https://codeload.github.com/huoyijie/AdvancedEAST/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":246668901,"owners_count":20814744,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["advancedeast","advancedeast-network-arch","algorithm","bellow","computer-vision","deep-learning","east","icpr","keras","machine-learning","python","scene","tensorflow","text-detect","text-predictions","tian-chi","tianchi"],"created_at":"2024-08-01T11:00:26.963Z","updated_at":"2025-04-02T12:31:48.777Z","avatar_url":"https://github.com/huoyijie.png","language":"Python","funding_links":[],"categories":["Text detection and localization","Python"],"sub_categories":["EAST [paper:2017](https://arxiv.org/abs/1704.03155)"],"readme":"# AdvancedEAST\nAdvancedEAST is an algorithm used for Scene image text detect,\nwhich is primarily based on\n[EAST:An Efficient and Accurate Scene Text Detector](https://arxiv.org/abs/1704.03155v2),\nand the significant improvement was also made,\nwhich make long text predictions more accurate.\nIf this project is helpful to you, welcome to star.\nAnd if you have any problem, please contact me.\n* email:yijie.huo@foxmail.com\n* website:[https://huoyijie.cn](https://huoyijie.cn)\n\n# advantages\n* writen in keras, easy to read and run\n* base on EAST, an advanced text detect algorithm\n* easy to train the model\n* significant improvement was made, long text predictions more accurate.(please\nsee 'demo results' part bellow,\nand pay attention to the activation image,\nwhich starts with yellow grids, and ends with green grids.) \n\nIn my experiments,\nAdvancedEast has obtained much better prediction accuracy then East,\nespecially on long text. Since East calculates final vertexes coordinates with\nweighted mean values of predicted vertexes coordinates of all pixels. It is too\ndifficult to predict the 2 vertexes from the other side of the quadrangle.\nSee East limitations picked from original paper bellow.\n![East limitations](image/East.limitations.png \"East limitations\")\n\n# project files\n* config file:cfg.py,control parameters\n* pre-process data:\n    preprocess.py,resize image\n* label data:\n    label.py,produce label info\n* define network\n    network.py\n* define loss function\n    losses.py\n* execute training\n    advanced_east.py and data_generator.py\n* predict\n    predict.py and nms.py\n    \n**后置处理过程说明参见\n[后置处理(含原理图)](https://huoyijie.cn/blog/82c8e470-7562-11ea-98d3-6d733527e90f/play)**\n\n# network arch\n* AdvancedEast\n\n![AdvancedEast network arch](image/AdvancedEast.network.png \"AdvancedEast network arch\")\n\n**网络输出说明：\n输出层分别是1位score map, 是否在文本框内；2位vertex code，是否属于文本框边界像素以及是头还是尾；4位geo，是边界像素可以预测的2个顶点坐标。所有像素构成了文本框形状，然后只用边界像素去预测回归顶点坐标。边界像素定义为黄色和绿色框内部所有像素，是用所有的边界像素预测值的加权平均来预测头或尾的短边两端的两个顶点。头和尾部分边界像素分别预测2个顶点，最后得到4个顶点坐标。**\n\n[原理简介(含原理图)](https://huoyijie.cn/blog/9a37ea00-755f-11ea-98d3-6d733527e90f/play)\n\n* East\n\n![East network arch](image/East.network.png \"East network arch\")\n\n\n# setup\n* python 3.6.3+\n* tensorflow-gpu 1.5.0+(or tensorflow 1.5.0+)\n* keras 2.1.4+\n* numpy 1.14.1+\n* tqdm 4.19.7+\n\n# training\n* tianchi ICPR dataset download\n链接: https://pan.baidu.com/s/1NSyc-cHKV3IwDo6qojIrKA 密码: ye9y\n\n* prepare training data:make data root dir(icpr),\ncopy images to root dir, and copy txts to root dir,\ndata format details could refer to 'ICPR MTWI 2018 挑战赛二：网络图像的文本检测',\n[Link](https://tianchi.aliyun.com/competition/introduction.htm?spm=5176.100066.0.0.3bcad780oQ9Ce4\u0026raceId=231651)\n* modify config params in cfg.py, see default values.\n* python preprocess.py, resize image to 256*256,384*384,512*512,640*640,736*736,\nand train respectively could speed up training process.\n* python label.py\n* python advanced_east.py, train entrance\n* python predict.py -p demo/001.png, to predict\n* pretrain model download(use for test)\n链接: https://pan.baidu.com/s/1KO7tR_MW767ggmbTjIJpuQ 密码: kpm2\n\n# demo results\n![001原图](demo/001.png \"001原图\")\n![001激活图](demo/001.png_act.jpg \"001激活图\")\n![001预测图](demo/001.png_predict.jpg \"001预测图\")\n\n![004原图](demo/004.jpg \"004原图\")\n![004激活图](demo/004.jpg_act.jpg \"004激活图\")\n![004预测图](demo/004.jpg_predict.jpg \"004预测图\")\n\n![005原图](demo/005.png \"005原图\")\n![005激活图](demo/005.png_act.jpg \"005激活图\")\n![005预测图](demo/005.png_predict.jpg \"005预测图\")\n\n* compared with east based on vgg16\n\nAs you can see, although the text area prediction is very accurate, the vertex coordinates are not accurate enough.\n\n![001激活图](demo/001.png_act_east.jpg \"001激活图\")\n![001预测图](demo/001.png_predict_east.jpg \"001预测图\")\n\n# License\nThe codes are released under the MIT License.\n\n# references\n* [EAST:An Efficient and Accurate Scene Text Detector](https://arxiv.org/abs/1704.03155v2)\n\n* [CTPN:Detecting Text in Natural Image with Connectionist Text Proposal Network](https://arxiv.org/abs/1609.03605)\n\n* [Deep Matching Prior Network: Toward Tighter Multi-oriented Text Detection](https://arxiv.org/abs/1703.01425)\n\n\n**网络输出说明：\n输出层分别是1位score map, 是否在文本框内；2位vertex code，是否属于文本框边界像素以及是头还是尾；4位geo，是边界像素可以预测的2个顶点坐标。所有像素构成了文本框形状，然后只用边界像素去预测回归顶点坐标。边界像素定义为黄色和绿色框内部所有像素，是用所有的边界像素预测值的加权平均来预测头或尾的短边两端的两个顶点。头和尾部分边界像素分别预测2个顶点，最后得到4个顶点坐标。**\n\n[原理简介(含原理图)](https://huoyijie.cn/blog/9a37ea00-755f-11ea-98d3-6d733527e90f/play)\n\n**后置处理过程说明参见\n[后置处理(含原理图)](https://huoyijie.cn/blog/82c8e470-7562-11ea-98d3-6d733527e90f/play)**\n\n[A Simple RaspberryPi Car Project](https://github.com/huoyijie/raspberrypi-car)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fhuoyijie%2FAdvancedEAST","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fhuoyijie%2FAdvancedEAST","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fhuoyijie%2FAdvancedEAST/lists"}