[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-MaybeShewill-CV--lanenet-lane-detection":3,"tool-MaybeShewill-CV--lanenet-lane-detection":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":79,"owner_email":80,"owner_twitter":81,"owner_website":82,"owner_url":83,"languages":84,"stars":101,"forks":102,"last_commit_at":103,"license":104,"difficulty_score":10,"env_os":105,"env_gpu":106,"env_ram":107,"env_deps":108,"category_tags":113,"github_topics":114,"view_count":10,"oss_zip_url":81,"oss_zip_packed_at":81,"status":16,"created_at":123,"updated_at":124,"faqs":125,"releases":156},1103,"MaybeShewill-CV\u002Flanenet-lane-detection","lanenet-lane-detection","Unofficial implemention of lanenet model for real time lane detection","Lanenet-Lane-Detection 是一个基于深度学习的实时车道线检测工具，通过TensorFlow框架实现，主要应用于自动驾驶和智能交通场景。该工具结合编码器-解码器结构与实例分割技术，利用Discriminative Loss Function提升检测精度，能够在复杂路况下高效识别车道线边界。其核心优势在于支持实时处理，单张图片推理速度可达50帧\u002F秒，且提供预训练模型供直接使用。\n\n该工具解决了传统车道线检测中精度不足、计算效率低等问题，尤其适用于需要快速响应的场景。开发者可通过自定义数据集训练模型，适应不同道路环境。对于研究人员，其开源代码和详细文档提供了良好的实验基础；对于自动驾驶开发者，预训练模型和测试脚本可快速集成到系统中。\n\n工具采用双阶段分割策略（二值分割+实例分割），在保持高精度的同时降低计算负担，适合需要平衡速度与准确性的应用。目前支持Tusimple数据集训练与测试，用户可直接下载预训练模型并进行本地验证，或通过脚本生成自定义数据集进行模型优化。","# LaneNet-Lane-Detection\nUse tensorflow to implement a Deep Neural Network for real time lane detection mainly based on the IEEE IV conference \npaper \"Towards End-to-End Lane Detection: an Instance Segmentation Approach\".You can refer to their paper for details \nhttps:\u002F\u002Farxiv.org\u002Fabs\u002F1802.05591. This model consists of a encoder-decoder stage, binary semantic segmentation stage \nand instance semantic segmentation using discriminative loss function for real time lane detection task.\n\nThe main network architecture is as follows:\n\n`Network Architecture`\n![NetWork_Architecture](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_9a8a9d742e54.png)\n\n## Installation\nThis software has only been tested on ubuntu 16.04(x64), python3.5, cuda-9.0, cudnn-7.0 with a GTX-1070 GPU. \nTo install this software you need tensorflow 1.12.0 and other version of tensorflow has not been tested but I think \nit will be able to work properly in tensorflow above version 1.12. Other required package you may install them by\n\n```\npip3 install -r requirements.txt\n```\n\n## Test model\nIn this repo I uploaded a model trained on tusimple lane dataset [Tusimple_Lane_Detection](http:\u002F\u002Fbenchmark.tusimple.ai\u002F#\u002F).\nThe deep neural network inference part can achieve around a 50fps which is similar to the description in the paper. But\nthe input pipeline I implemented now need to be improved to achieve a real time lane detection system.\n\nThe trained lanenet model weights files are stored in \n[lanenet_pretrained_model](https:\u002F\u002Fwww.dropbox.com\u002Fsh\u002F0b6r0ljqi76kyg9\u002FAADedYWO3bnx4PhK1BmbJkJKa?dl=0). You can \ndownload the model and put them in folder weights\u002Ftusimple_lanenet\u002F\n\nYou may also download the pretrained model via [BaiduNetDisk here](https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A) and\nextract code is `86sd`.\n\nYou can test a single image on the trained model as follows\n\n```\npython tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--image_path https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_107befcdf558.jpg\n```\nThe results are as follows:\n\n`Test Input Image`\n\n![Test Input](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_107befcdf558.jpg)\n\n`Test Lane Mask Image`\n\n![Test Lane_Mask](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_33c5c67168be.png)\n\n`Test Lane Binary Segmentation Image`\n\n![Test Lane_Binary_Seg](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b0ef0a64a555.png)\n\n`Test Lane Instance Segmentation Image`\n\n![Test Lane_Instance_Seg](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_ea0d47cf8cab.png)\n\nIf you want to evaluate the model on the whole tusimple test dataset you may call\n```\npython tools\u002Fevaluate_lanenet_on_tusimple.py \n--image_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Fclips \n--weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--save_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Ftest_output\n```\nIf you set the save_dir argument the result will be saved in that folder \nor the result will not be saved but be \ndisplayed during the inference process holding on 3 seconds per image. \nI test the model on the whole tusimple lane \ndetection dataset and make it a video. You may catch a glimpse of it bellow.\n\n`Tusimple test dataset gif`\n![tusimple_batch_test_gif](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_f74257efd3e8.gif)\n\n## Train your own model\n#### Data Preparation\nFirstly you need to organize your training data refer to the data\u002Ftraining_data_example folder structure. And you need \nto generate a train.txt and a val.txt to record the data used for training the model. \n\nThe training samples consist of three components, a binary segmentation label file, a instance segmentation label\nfile and the original image. The binary segmentation uses 255 to represent the lane field and 0 for the rest. The \ninstance use different pixel value to represent different lane field and 0 for the rest.\n\nAll your training image will be scaled into the same scale according to the config file.\n\nUse the script here to generate the tensorflow records file\n\n```\npython tools\u002Fmake_tusimple_tfrecords.py \n```\n\n#### Train model\nIn my experiment the training epochs are 80010, batch size is 4, initialized learning rate is 0.001 and use polynomial \ndecay with power 0.9. About training parameters you can check the global_configuration\u002Fconfig.py for details. \nYou can switch --net argument to change the base encoder stage. If you choose --net vgg then the vgg16 will be used as \nthe base encoder stage and a pretrained parameters will be loaded. And you can modified the training \nscript to load your own pretrained parameters or you can implement your own base encoder stage. \nYou may call the following script to train your own model\n\n```\npython tools\u002Ftrain_lanenet_tusimple.py \n```\n\nYou may monitor the training process using tensorboard tools\n\nDuring my experiment the `Total loss` drops as follows:  \n![Training loss](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_cbdbb2f7f13a.png)\n\nThe `Binary Segmentation loss` drops as follows:  \n![Training binary_seg_loss](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_0c06d04b2f76.png)\n\nThe `Instance Segmentation loss` drops as follows:  \n![Training instance_seg_loss](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_476b0d58dabe.png)\n\n## Experiment\nThe accuracy during training process rises as follows: \n![Training accuracy](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_c7bd5fca3ac0.png)\n\nPlease cite my repo [lanenet-lane-detection](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection) if you use it.\n\n## Segment-Anything-U-Specify\n\nYou must be interested in recently released SAM model. Here's a repo using clip + sam to segment any instances you specify.\n[segment-anything-u-specify](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Fsegment-anything-u-specify).\n\n\u003Cp align=\"left\">\n  \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_5602201b2f31.png' alt='segment-anything-u-specify'>\n\u003C\u002Fp>\n\n## Serve Your Model\n\nIf you want to serve your model via a web server you may be interested in [mortred_model_server](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Fmortred_model_server) which is a high performace web server for DNN vision models  :fire::fire::fire:\n\n\u003Cp align=\"left\">\n  \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b9d394a72b19.png' alt='mortred_model_server'>\n\u003C\u002Fp>\n\n## Recently updates 2018.11.10\nAdjust some basic cnn op according to the new tensorflow api. Use the \ntraditional SGD optimizer to optimize the whole model instead of the\norigin Adam optimizer used in the origin paper. I have found that the\nSGD optimizer will lead to more stable training process and will not \neasily stuck into nan loss which may often happen when using the origin\ncode.\n\n## Recently updates 2018.12.13\nSince a lot of user want a automatic tools to generate the training samples\nfrom the Tusimple Dataset. I upload the tools I use to generate the training\nsamples. You need to firstly download the Tusimple dataset and unzip the \nfile to your local disk. Then run the following command to generate the \ntraining samples and the train.txt file.\n\n```angular2html\npython tools\u002Fgenerate_tusimple_dataset.py --src_dir path\u002Fto\u002Fyour\u002Funzipped\u002Ffile\n```\n\nThe script will make the train folder and the test folder. The training \nsamples of origin rgb image, binary label image, instance label image will\nbe automatically generated in the training\u002Fgt_image, training\u002Fgt_binary_image,\ntraining\u002Fgt_instance_image folder.You may check it yourself before start\nthe training process.\n\nPay attention that the script only process the training samples and you \nneed to select several lines from the train.txt to generate your own \nval.txt file. In order to obtain the test images you can modify the \nscript on your own.\n\n## Recently updates 2020.06.12\n\nAdd real-time segmentation model BiseNetV2 as lanenet backbone. You may modify the\nconfig\u002Ftusimple_lanenet.yaml config file to choose the front-end of lanenet model.\n\nNew lanenet model trainned based on BiseNetV2 can be found [here](https:\u002F\u002Fwww.dropbox.com\u002Fsh\u002F0b6r0ljqi76kyg9\u002FAADedYWO3bnx4PhK1BmbJkJKa?dl=0)\n\n[BaiduNetDisk](https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A) is available too.\nYou can download here https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A and extract code is `86sd`\n\nThe new model can reach 78 fps in single image inference process.\n\n## Recently updates 2022.05.28\n\nSince lots of user have encountered with a empty mask image problem when they do model inference using their own custom\ndata. For example the user [issue](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F382) have encountered\nsuch a problem. I have openend a discussion [here](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fdiscussions\u002F561#discussion-4104802)\nto give some advice to solve those problem.\n\nThat problem mainly caused by the dbscan cluster's params was not properly adjusted for custom data. For example if I use\nthe default dbscan param settled [here](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fblob\u002F5f704c86759b0b65955fb27c85a42f343c1c8c5c\u002Fconfig\u002Ftusimple_lanenet.yaml#L90-L93)\n```\nPOSTPROCESS:\n    MIN_AREA_THRESHOLD: 100\n    DBSCAN_EPS: 0.35\n    DBSCAN_MIN_SAMPLES: 1000\n```\nThe inference result was\n![black_mask](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_41b9ba9f0698.png)\n\nWhen I enlarge the dbscan DBSCAN_EPS param from 0.35 to 0.5 and reduce DBSCAN_MIN_SAMPLES from 1000 to 250. The infer\nence result was\n![black_mask_after_adjust](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_bedb9ee14c59.png)\n\nSome more detailed discussion you may find in [discussion module](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fdiscussions\u002F561#discussion-4104802)\n\nThe lane fit process in postprocess module was designed for tusimple dataset which means it can not work well on your\ncustorm data. So I add an option in testing scripts to disable this feature when processing custom data. It will plot\nmask image directly upon source image\n\n```\npython tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--image_path .\u002Fdata\u002Fcustom_test_image\u002Ftest.png --with_lane_fit 0\n```\n\nBefore you test the example custom data remember to adjust dbscan cluster params following instruction above and the test\nresult should be like\n![black_mask_after_adjust](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_bedb9ee14c59.png)\n\nTo get better lane detection result on your own data you'd better train a new model on custom dataset rather than\nusing the pretrained model directly.\n\nHope it helps:)\n\n## MNN Project\n\nAdd tools to convert lanenet tensorflow ckpt model into mnn model and deploy\nthe model on mobile device\n\n#### Freeze your tensorflow ckpt model weights file\n```\ncd LANENET_PROJECT_ROOT_DIR\npython mnn_project\u002Ffreeze_lanenet_model.py -w lanenet.ckpt -s lanenet.pb\n```\n\n#### Convert pb model into mnn model\n```\ncd MNN_PROJECT_ROOT_DIR\u002Ftools\u002Fconverter\u002Fbuild\n.\u002FMNNConver -f TF --modelFile lanenet.pb --MNNModel lanenet.mnn --bizCode MNN\n```\n\n#### Add lanenet source code into MNN project \n\nAdd lanenet source code into MNN project and modified CMakeList.txt to \ncompile the executable binary file.\n\n## Status\n\n![Repobeats analytics image](https:\u002F\u002Frepobeats.axiom.co\u002Fapi\u002Fembed\u002Fffa5169a3a4002d4f573b12be173c9382d14b78a.svg \"Repobeats analytics image\")\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b75651909de9.png)](https:\u002F\u002Fstar-history.com\u002F#MaybeShewill-CV\u002Flanenet-lane-detection&Date)\n\n## TODO\n- [x] Add a embedding visualization tools to visualize the embedding feature map\n- [x] Add detailed explanation of training the components of lanenet separately.\n- [x] Training the model on different dataset\n- ~~[ ] Adjust the lanenet hnet model and merge the hnet model to the main lanenet model~~\n- ~~[ ] Change the normalization function from BN to GN~~\n\n## Acknowledgement\n\nThe lanenet project refers to the following projects:\n\n- [MNN](https:\u002F\u002Fgithub.com\u002Falibaba\u002FMNN)\n- [SimpleDBSCAN](https:\u002F\u002Fgithub.com\u002FCallmeNezha\u002FSimpleDBSCAN)\n- [PaddleSeg](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleSeg)\n\n## Visitor Count\n\n![Visitor Count](https:\u002F\u002Fprofile-counter.glitch.me\u002F15725187_lanenet\u002Fcount.svg)\n\n## Contact\n\nScan the following QR to disscuss :)\n![qr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_e85f1fb91fb4.jpg)\n","# LaneNet-车道检测\n使用 tensorflow 实现一个深度神经网络用于实时车道检测，主要基于 IEEE IV 会议论文 \"Towards End-to-End Lane Detection: an Instance Segmentation Approach\"。详情请参考其论文 https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.05591。该模型包含编码器-解码器阶段、二元语义分割阶段和使用判别损失函数的实例语义分割阶段，用于实时车道检测任务。\n\n主要网络架构如下：\n\n`网络架构`\n![NetWork_Architecture](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_9a8a9d742e54.png)\n\n## 安装\n该软件仅在 ubuntu 16.04(x64)、python3.5、cuda-9.0、cudnn-7.0 配置下测试通过，配备 GTX-1070 GPU。安装此软件需要 tensorflow 1.12.0，其他版本的 tensorflow 尚未测试，但我认为在 tensorflow 1.12 版本以上应该可以正常工作。其他所需包可通过以下命令安装：\n\n```\npip3 install -r requirements.txt\n```\n\n## 测试模型\n本仓库上传了一个在 tusimple 车道数据集 [Tusimple_Lane_Detection](http:\u002F\u002Fbenchmark.tusimple.ai\u002F#\u002F ) 上训练的模型。深度神经网络推理部分可以达到约 50fps，与论文中的描述相似。但目前实现的输入管道需要改进以实现真正的实时车道检测系统。\n\n训练好的 lanenet 模型权重文件存储在 \n[lanenet_pretrained_model](https:\u002F\u002Fwww.dropbox.com\u002Fsh\u002F0b6r0ljqi76kyg9\u002FAADedYWO3bnx4PhK1BmbJkJKa?dl=0)。您可以下载模型并将它们放入 weights\u002Ftusimple_lanenet\u002F 文件夹中。\n\n您也可以通过 [BaiduNetDisk 这里](https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A) 下载预训练模型，提取代码为 `86sd`。\n\n您可以使用以下命令在训练模型上测试单张图像：\n\n```\npython tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--image_path https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_107befcdf558.jpg\n```\n结果如下：\n\n`测试输入图像`\n\n![测试输入](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_107befcdf558.jpg)\n\n`测试车道掩码图像`\n\n![测试 Lane_Mask](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_33c5c67168be.png)\n\n`测试车道二元语义分割图像`\n\n![测试 Lane_Binary_Seg](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b0ef0a64a555.png)\n\n`测试车道实例语义分割图像`\n\n![测试 Lane_Instance_Seg](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_ea0d47cf8cab.png)\n\n如果您想在整个 tusimple 测试数据集上评估模型，可以调用\n```\npython tools\u002Fevaluate_lanenet_on_tusimple.py \n--image_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Fclips \n--weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--save_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Ftest_output\n```\n如果设置了 save_dir 参数，结果将保存在该文件夹中，否则在推理过程中会显示结果并每张图像保持 3 秒钟。我已在整个 tusimple 车道检测数据集上测试了该模型并制作成视频。您可下方查看。\n\n`Tusimple 测试数据集 gif`\n![tusimple_batch_test_gif](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_f74257efd3e8.gif)\n\n## 训练自己的模型\n#### 数据准备\n首先您需要按照 data\u002Ftraining_data_example 文件夹结构组织训练数据。还需要生成 train.txt 和 val.txt 来记录用于训练模型的数据。\n\n训练样本包含三个组成部分：二元语义分割标签文件、实例语义分割标签文件和原始图像。二元语义分割使用 255 表示车道区域，0 表示其余部分。实例语义分割使用不同像素值表示不同车道区域，0 表示其余部分。\n\n所有训练图像都将根据配置文件缩放至相同尺度。\n\n使用此处的脚本生成 tensorflow 记录文件\n\n```\npython tools\u002Fmake_tusimple_tfrecords.py \n```\n\n#### 训练模型\n在我的实验中，训练轮数为 80010，批量大小为 4，初始学习率为 0.001，使用多项式衰减（power 0.9）。关于训练参数详情请查看 global_configuration\u002Fconfig.py。您可以通过切换 --net 参数更改基础编码器阶段。如果您选择 --net vgg，则使用 vgg16 作为基础编码器阶段并加载预训练参数。您还可以修改训练脚本加载自己的预训练参数，或实现自己的基础编码器阶段。您可以调用以下脚本训练自己的模型\n\n```\npython tools\u002Ftrain_lanenet_tusimple.py \n```\n\n您可以使用 tensorboard 工具监控训练过程\n\n在我的实验中，`总损失` 下降情况如下：  \n![训练损失](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_cbdbb2f7f13a.png)\n\n`二元语义分割损失` 下降情况如下：  \n![训练 binary_seg_loss](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_0c06d04b2f76.png)\n\n`实例语义分割损失` 下降情况如下：  \n![训练 instance_seg_loss](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_476b0d58dabe.png)\n\n## 实验\n训练过程中的准确率提升情况如下： \n![训练准确率](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_c7bd5fca3ac0.png)\n\n如您使用本仓库，请引用我的 repo [lanenet-lane-detection](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection)。\n\n## Segment-Anything-U-Specify\n\n您可能对最近发布的 SAM 模型感兴趣。这里有一个使用 clip + sam 实现的仓库，可用于指定分割任意实例。\n[segment-anything-u-specify](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Fsegment-anything-u-specify).\n\n\u003Cp align=\"left\">\n  \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_5602201b2f31.png' alt='segment-anything-u-specify'>\n\u003C\u002Fp>\n\n## 服务您的模型\n\n如果您希望通过 Web 服务器服务您的模型，可能对 [mortred_model_server](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Fmortred_model_server) 感兴趣，这是一个高性能的 DNN 视觉模型 Web 服务器 :fire::fire::fire:\n\n\u003Cp align=\"left\">\n  \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b9d394a72b19.png' alt='mortred_model_server'>\n\u003C\u002Fp>\n\n## 最近更新 2018.11.10\n根据新的 tensorflow API 调整了一些基本的 CNN 操作。使用传统的 SGD 优化器优化整个模型，而不是原论文中使用的原始 Adam 优化器。我发现 SGD 优化器能带来更稳定的训练过程，并且不容易陷入 nan 损失，这在使用原始代码时经常发生。\n\n## 最近更新 2018.12.13  \n由于许多用户希望有一个自动工具来从Tusimple数据集生成训练样本，我上传了我用来生成训练样本的工具。您需要首先下载Tusimple数据集并将文件解压到本地磁盘。然后运行以下命令来生成训练样本和train.txt文件。\n\n```angular2html\npython tools\u002Fgenerate_tusimple_dataset.py --src_dir path\u002Fto\u002Fyour\u002Funzipped\u002Ffile\n```\n\n该脚本会创建train文件夹和test文件夹。原始rgb图像、二值标签图像、实例标签图像的训练样本将自动生成在training\u002Fgt_image、training\u002Fgt_binary_image、training\u002Fgt_instance_image文件夹中。您在开始训练过程前可以自行检查。\n\n请注意，该脚本仅处理训练样本，您需要从train.txt中选择几行来生成自己的val.txt文件。为了获取测试图像，您可以自行修改脚本。\n\n## 最近更新 2020.06.12  \n\n新增基于BiseNetV2的实时分割模型BiseNetV2作为lanenet的backbone。您可以通过修改config\u002Ftusimple_lanenet.yaml配置文件来选择lanenet模型的前端。\n\n基于BiseNetV2训练的新lanenet模型可在此处找到 [here](https:\u002F\u002Fwww.dropbox.com\u002Fsh\u002F0b6r0ljqi76kyg9\u002FAADedYWO3bnx4PhK1BmbJkJKa?dl=0)\n\n[BaiduNetDisk](https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A) 也提供。您可在此处下载 https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A 并使用提取码 `86sd`\n\n新模型在单张图像推理过程中可达78 fps。\n\n## 最近更新 2022.05.28  \n\n由于许多用户在使用自定义数据进行模型推理时遇到了空mask图像问题。例如用户 [issue](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F382) 遇到了此类问题。我在 [此处](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fdiscussions\u002F561#discussion-4104802) 开启了讨论以提供解决建议。\n\n该问题主要由dbscan聚类算法参数未针对自定义数据进行适当调整导致。例如如果我使用默认的dbscan参数设定 [此处](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fblob\u002F5f704c86759b0b65955fb27c85a42f343c1c8c5c\u002Fconfig\u002Ftusimple_lanenet.yaml#L90-L93)\n```\nPOSTPROCESS:\n    MIN_AREA_THRESHOLD: 100\n    DBSCAN_EPS: 0.35\n    DBSCAN_MIN_SAMPLES: 1000\n```\n推理结果为\n![black_mask](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_41b9ba9f0698.png)\n\n当我将dbscan DBSCAN_EPS参数从0.35增加到0.5，并将DBSCAN_MIN_SAMPLES参数从1000减少到250时，推理结果为\n![black_mask_after_adjust](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_bedb9ee14c59.png)\n\n更多详细讨论您可参考 [讨论模块](https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fdiscussions\u002F561#discussion-4104802)\n\npostprocess模块中的车道拟合过程是为Tusimple数据集设计的，这意味着它在您的自定义数据上无法很好地工作。因此我在测试脚本中添加了一个选项，在处理自定义数据时禁用此功能。它会在源图像上直接绘制mask图像\n\n```\npython tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \n--image_path .\u002Fdata\u002Fcustom_test_image\u002Ftest.png --with_lane_fit 0\n```\n\n在测试示例自定义数据前，请按照上述说明调整dbscan聚类参数，测试结果应类似\n![black_mask_after_adjust](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_bedb9ee14c59.png)\n\n为了在您的数据上获得更好的车道检测结果，最好在自定义数据集上训练新模型，而不是直接使用预训练模型。\n\n希望有所帮助:)\n\n## MNN项目\n\n添加工具将lanenet tensorflow ckpt模型转换为mnn模型并在移动设备上部署\n\n#### 冻结tensorflow ckpt模型权重文件\n```\ncd LANENET_PROJECT_ROOT_DIR\npython mnn_project\u002Ffreeze_lanenet_model.py -w lanenet.ckpt -s lanenet.pb\n```\n\n#### 将pb模型转换为mnn模型\n```\ncd MNN_PROJECT_ROOT_DIR\u002Ftools\u002Fconverter\u002Fbuild\n.\u002FMNNConver -f TF --modelFile lanenet.pb --MNNModel lanenet.mnn --bizCode MNN\n```\n\n#### 将lanenet源代码添加到MNN项目 \n\n将lanenet源代码添加到MNN项目并修改CMakeList.txt以编译可执行二进制文件。\n\n## 状态\n\n![Repobeats分析图像](https:\u002F\u002Frepobeats.axiom.co\u002Fapi\u002Fembed\u002Fffa5169a3a4002d4f573b12be173c9382d14b78a.svg \"Repobeats分析图像\")\n\n## 星星历史\n\n[![星星历史图表](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_b75651909de9.png)](https:\u002F\u002Fstar-history.com\u002F#MaybeShewill-CV\u002Flanenet-lane-detection&Date)\n\n## 待办事项\n- [x] 添加嵌入可视化工具以可视化嵌入特征图\n- [x] 添加详细解释如何单独训练lanenet的各个组件。\n- [x] 在不同数据集上训练模型\n- ~~[ ] 调整lanenet hnet模型并将hnet模型合并到主lanenet模型~~\n- ~~[ ] 将归一化函数从BN改为GN~~\n\n## 致谢\n\nlanenet项目参考了以下项目：\n\n- [MNN](https:\u002F\u002Fgithub.com\u002Falibaba\u002FMNN)\n- [SimpleDBSCAN](https:\u002F\u002Fgithub.com\u002FCallmeNezha\u002FSimpleDBSCAN)\n- [PaddleSeg](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleSeg)\n\n## 访问次数\n\n![访问次数](https:\u002F\u002Fprofile-counter.glitch.me\u002F15725187_lanenet\u002Fcount.svg)\n\n## 联系方式\n\n扫描以下二维码讨论:)\n![qr](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_readme_e85f1fb91fb4.jpg)","# lanenet-lane-detection 快速上手指南\n\n## 环境准备\n- **系统要求**：Ubuntu 16.04（x64）、Python 3.5、CUDA 9.0、cudnn 7.0、GTX-1070 GPU\n- **前置依赖**：TensorFlow 1.12.0（推荐版本）  \n  其他依赖通过以下命令安装：\n  ```bash\n  pip3 install -r requirements.txt\n  ```\n\n## 安装步骤\n1. 下载预训练模型权重：\n   - Dropbox链接：[lanenet_pretrained_model](https:\u002F\u002Fwww.dropbox.com\u002Fsh\u002F0b6r0ljqi76kyg9\u002FAADedYWO3bnx4PhK1BmbJkJKa?dl=0)\n   - 百度网盘：[下载地址](https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1sLLSE1CWksKNxmRIGaQn_A)（提取码：86sd）\n   - 将模型文件放入 `weights\u002Ftusimple_lanenet\u002F` 文件夹\n\n2. 测试模型：\n   ```bash\n   python tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \\\n                               --image_path .\u002Fdata\u002Ftusimple_test_image\u002F0.jpg\n   ```\n\n## 基本使用\n- **单图测试**：\n  ```bash\n  python tools\u002Ftest_lanenet.py --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \\\n                              --image_path .\u002Fdata\u002Ftusimple_test_image\u002F0.jpg\n  ```\n  输出结果包含车道线二值分割、实例分割等多通道图像\n\n- **全数据集评估**：\n  ```bash\n  python tools\u002Fevaluate_lanenet_on_tusimple.py \\\n    --image_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Fclips \\\n    --weights_path \u002FPATH\u002FTO\u002FYOUR\u002FCKPT_FILE_PATH \\\n    --save_dir ROOT_DIR\u002FTUSIMPLE_DATASET\u002Ftest_set\u002Ftest_output\n  ```","自动驾驶初创公司「智行科技」的算法团队正在开发L2级辅助驾驶系统，需要在嵌入式设备上实现全天候车道线检测。传统CV算法在复杂路况中频繁失效，团队尝试部署深度学习模型时面临实时性和准确率的双重挑战。\n\n### 没有 lanenet-lane-detection 时\n- 夜间逆光场景下，传统霍夫变换算法漏检率达40%，导致车道保持功能失效\n- 雨雪天气需手动调整10+个参数，每次调参耗时2-3小时且效果不稳定\n- 基于ResNet的检测模型推理延迟达800ms，无法满足ADAS系统实时性要求\n- 多车道场景中无法区分相邻车道线，导致变道决策错误率提升25%\n- 自建数据集需从零训练模型，预训练模型迁移效果差，迭代周期长达3周\n\n### 使用 lanenet-lane-detection 后\n- 二值分割与实例分割双任务架构使夜间检测准确率提升至92%，逆光场景漏检率降至8%\n- 端到端训练流程自动优化特征提取，参数调整工作量减少70%，单次调优仅需30分钟\n- 轻量化网络结构实现50FPS实时推理，端到端延迟压缩至20ms内\n- 实例分割输出支持4车道并行识别，变道决策准确率提升至95%\n- 预训练模型在自建数据集微调仅需3天即可收敛，mIOU指标达到0.87\n\n**核心价值**：lanenet-lane-detection通过端到端深度学习架构与优化的推理速度，在复杂工况下实现亚像素级车道线检测，使自动驾驶系统的车道感知模块达到车规级可靠性要求。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FMaybeShewill-CV_lanenet-lane-detection_33c5c671.png","MaybeShewill-CV","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FMaybeShewill-CV_63d894bb.png","Computer Vision R&D","Baidu","Tong Ji University","howard327@163.com",null,"https:\u002F\u002Fmaybeshewill-cv.github.io","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV",[85,89,93,97],{"name":86,"color":87,"percentage":88},"Python","#3572A5",78,{"name":90,"color":91,"percentage":92},"C++","#f34b7d",20,{"name":94,"color":95,"percentage":96},"C","#555555",1.9,{"name":98,"color":99,"percentage":100},"Shell","#89e051",0.1,2539,904,"2026-04-02T08:46:21","Apache-2.0","Linux","需要 NVIDIA GPU，显存 8GB+，CUDA 9.0","未说明",{"notes":109,"python":110,"dependencies":111},"需下载预训练模型（约500MB），调整DBSCAN参数适配自定义数据。测试时可通过--with_lane_fit 0禁用车道拟合功能。建议使用GTX 1070或更高性能显卡实现实时检测。","3.5",[112],"tensorflow>=1.12.0",[14,13],[115,116,117,118,119,120,121,122],"lanenet","self-driving-car","deep-learning","instance-segmentation","tensorflow","lane-detection","lane-finding","lane-lines-detection","2026-03-27T02:49:30.150509","2026-04-06T07:14:01.665055",[126,131,136,141,146,151],{"id":127,"question_zh":128,"answer_zh":129,"source_url":130},4965,"训练时出现loss为nan的问题如何解决？","将代码中的tf.norm替换为tf.reduce_sum(tf.square())，并调整cutoff variance distance和cutoff cluster distance参数。同时检查TensorFlow版本是否为1.15.0，确保数据预处理和路径配置正确。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F33",{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},4966,"训练过程中出现nan值的原因是什么？","可能是由于计算过程中出现除零错误或数值不稳定。建议检查数据预处理是否正常，确保输入数据范围合理，并尝试降低学习率或调整模型参数以避免数值溢出。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F116",{"id":137,"question_zh":138,"answer_zh":139,"source_url":140},4967,"如何解决自定义数据集训练时的reshape错误？","确保生成的JSON文件格式正确，标注坐标需与TuSimple数据集一致。使用Matlab工具标注车道线坐标后，通过脚本生成符合要求的JSON文件，保证图像路径和标注数据的对应关系。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F343",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},4968,"重新训练模型时出现OpenCV错误如何处理？","检查二值图和实例图的文件路径是否正确，确保训练数据格式符合要求。若使用自定义数据集，需确认图像尺寸与模型输入匹配，并验证train.txt和val.txt文件中的路径是否正确。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F74",{"id":147,"question_zh":148,"answer_zh":149,"source_url":150},4969,"Embedding branch模块的代码是否存在异常？","论文中提到的Embedding branch模块已包含在代码中，可通过mean shift聚类方法实现。若遇到异常，请检查代码中是否正确实现了论文中的特征聚合逻辑，并确保输入数据格式符合要求。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F67",{"id":152,"question_zh":153,"answer_zh":154,"source_url":155},4970,"模型是否支持使用CULane数据集训练？","当前模型主要基于TuSimple数据集训练，未直接支持CULane数据集。若需使用CULane数据集，需确保标注格式与TuSimple一致，并验证数据预处理步骤是否适配新数据集。","https:\u002F\u002Fgithub.com\u002FMaybeShewill-CV\u002Flanenet-lane-detection\u002Fissues\u002F401",[]]