{"id":13443173,"url":"https://github.com/PaddlePaddle/PaddleYOLO","last_synced_at":"2025-03-20T16:30:46.689Z","repository":{"id":49523463,"uuid":"513505410","full_name":"PaddlePaddle/PaddleYOLO","owner":"PaddlePaddle","description":"🚀🚀🚀 YOLO series of PaddlePaddle implementation, PP-YOLOE+, RT-DETR, YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLOv10, YOLO11, YOLOX, YOLOv5u, YOLOv7u, YOLOv6Lite, RTMDet and so on. 🚀🚀🚀","archived":false,"fork":false,"pushed_at":"2025-02-26T03:29:08.000Z","size":314391,"stargazers_count":585,"open_issues_count":16,"forks_count":141,"subscribers_count":15,"default_branch":"develop","last_synced_at":"2025-03-16T18:05:50.262Z","etag":null,"topics":["instance-segmentation","object-detection","paddleyolo","ppyolo","ppyoloe","ppyolov2","rt-detr","rtmdet","yolo","yolo11","yolov10","yolov3","yolov5","yolov5u","yolov6","yolov6lite","yolov7","yolov7u","yolov8","yolox"],"latest_commit_sha":null,"homepage":"https://github.com/PaddlePaddle/PaddleYOLO","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"gpl-3.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/PaddlePaddle.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,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-07-13T12:05:19.000Z","updated_at":"2025-03-13T16:10:12.000Z","dependencies_parsed_at":"2023-02-17T21:15:47.979Z","dependency_job_id":"a081c179-6021-4ad8-9480-fbf8d0ed0200","html_url":"https://github.com/PaddlePaddle/PaddleYOLO","commit_stats":{"total_commits":2121,"total_committers":133,"mean_commits":"15.947368421052632","dds":0.8330975954738331,"last_synced_commit":"e37222a2660e02b8779cae6dbc68efdc1bb7e56b"},"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PaddlePaddle%2FPaddleYOLO","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PaddlePaddle%2FPaddleYOLO/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PaddlePaddle%2FPaddleYOLO/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PaddlePaddle%2FPaddleYOLO/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/PaddlePaddle","download_url":"https://codeload.github.com/PaddlePaddle/PaddleYOLO/tar.gz/refs/heads/develop","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":244649707,"owners_count":20487471,"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":["instance-segmentation","object-detection","paddleyolo","ppyolo","ppyoloe","ppyolov2","rt-detr","rtmdet","yolo","yolo11","yolov10","yolov3","yolov5","yolov5u","yolov6","yolov6lite","yolov7","yolov7u","yolov8","yolox"],"created_at":"2024-07-31T03:01:57.071Z","updated_at":"2025-03-20T16:30:46.682Z","avatar_url":"https://github.com/PaddlePaddle.png","language":"Python","funding_links":[],"categories":["Summary","Python"],"sub_categories":[],"readme":"简体中文 | [English](README_en.md)\n\n## 简介\n\n**PaddleYOLO**是基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)的YOLO系列模型库，**只包含YOLO系列模型的相关代码**，支持`YOLOv3`、`PP-YOLO`、`PP-YOLOv2`、`PP-YOLOE`、**`PP-YOLOE+`**、**`RT-DETR`**、`YOLOX`、`YOLOv5`、`YOLOv6`、`YOLOv7`、`YOLOv8`、`YOLOv5u`、`YOLOv7u`、`YOLOv6Lite`、`RTMDet`等模型，COCO数据集模型库请参照 [ModelZoo](docs/MODEL_ZOO_cn.md) 和 [configs](configs/)。\n\n\u003cdiv  align=\"center\"\u003e\n  \u003cimg src=\"https://user-images.githubusercontent.com/13104100/213197403-c8257486-9ac4-486f-a0d5-4e3fe27ca852.jpg\" width=\"480\"/\u003e\n  \u003cimg src=\"https://user-images.githubusercontent.com/13104100/213197635-eeb55433-bb2d-44f6-b374-73c616cfab24.jpg\" width=\"480\"/\u003e\n\u003c/div\u003e\n\n**注意:**\n\n - **PaddleYOLO** 代码库协议为 **[GPL 3.0](LICENSE)**，[YOLOv5](configs/yolov5)、[YOLOv6](configs/yolov6)、[YOLOv7](configs/yolov7)和[YOLOv8](configs/yolov8)这几类模型代码不合入[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)，其余YOLO模型推荐在[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)中使用，**会最先发布PP-YOLO系列特色检测模型的最新进展**；\n - **PaddleYOLO**代码库**推荐使用paddlepaddle-2.4.2以上的版本**，请参考[官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)下载对应适合版本，**Windows平台请安装paddle develop版本**；\n - **PaddleYOLO 的[Roadmap](https://github.com/PaddlePaddle/PaddleYOLO/issues/44)** issue用于收集用户的需求，欢迎提出您的建议和需求；\n\n## 教程\n\n\u003cdetails open\u003e\n\u003csummary\u003e安装\u003c/summary\u003e\n\nClone 代码库和安装 [requirements.txt](./requirements.txt)，环境需要在一个\n[**Python\u003e=3.7.0**](https://www.python.org/) 下的环境，且需要安装\n[**PaddlePaddle\u003e=2.4.2**](https://www.paddlepaddle.org.cn/install/)。\n\n```bash\ngit clone https://github.com/PaddlePaddle/PaddleYOLO  # clone\ncd PaddleYOLO\npip install -r requirements.txt  # install\n```\n\n\u003c/details\u003e\n\n\u003cdetails open\u003e\n\u003csummary\u003e训练/验证/预测/\u003c/summary\u003e\n将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。\n\n```bash\nmodel_name=ppyoloe # 可修改，如 yolov7\njob_name=ppyoloe_plus_crn_s_80e_coco # 可修改，如 yolov7_tiny_300e_coco\n\nconfig=configs/${model_name}/${job_name}.yml\nlog_dir=log_dir/${job_name}\n# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams\nweights=output/${job_name}/model_final.pdparams\n\n# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练\n# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp\npython -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp\n\n# 2.评估，加 --classwise 表示输出每一类mAP\nCUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise\n\n# 3.预测 (单张图/图片文件夹）\nCUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5\n# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5\n```\n\n\u003c/details\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e部署/测速\u003c/summary\u003e\n\n将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。\n\n```bash\nmodel_name=ppyoloe # 可修改，如 yolov7\njob_name=ppyoloe_plus_crn_s_80e_coco # 可修改，如 yolov7_tiny_300e_coco\n\nconfig=configs/${model_name}/${job_name}.yml\nlog_dir=log_dir/${job_name}\n# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams\nweights=output/${job_name}/model_final.pdparams\n\n# 4.导出模型，以下3种模式选一种\n## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显\nCUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True\n\n## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度\n# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True\n\n## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度\n# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True\n\n# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测\nCUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU\n\n# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型\nCUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16\n\n# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型\npaddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx\n\n# 8.onnx trt测速\n/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16\n/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32\n```\n\n- 如果想切换模型，只要修改开头两行即可，如:\n  ```\n  model_name=yolov7\n  job_name=yolov7_tiny_300e_coco\n  ```\n- 导出**onnx**，首先安装[Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)，`pip install paddle2onnx`；\n- **统计FLOPs(G)和Params(M)**，首先安装[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)，`pip install paddleslim`，然后设置[runtime.yml](configs/runtime.yml)里`print_flops: True`和`print_params: True`，并且注意确保是**单尺度**下如640x640，**打印的是MACs，FLOPs=2*MACs**。\n\n\u003c/details\u003e\n\n\n\u003cdetails open\u003e\n\u003csummary\u003e [训练自定义数据集](https://github.com/PaddlePaddle/PaddleYOLO/issues/43) \u003c/summary\u003e\n\n- 请参照[文档](docs/MODEL_ZOO_cn.md#自定义数据集)和[issue](https://github.com/PaddlePaddle/PaddleYOLO/issues/43)；\n- PaddleDetection团队提供了**基于PP-YOLOE的各种垂类检测模型**的配置文件和权重，用户也可以作为参考去使用自定义数据集。请参考 [PP-YOLOE application](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppyoloe/application)、[pphuman](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/pphuman)、[ppvehicle](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppvehicle)、[visdrone](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/visdrone) 和 [smalldet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/smalldet)。\n- PaddleDetection团队也提供了**VOC数据集的各种YOLO模型**的配置文件和权重，用户也可以作为参考去使用自定义数据集。请参考 [voc](configs/voc)。\n- 训练自定义数据集之前请先**确保加载了对应COCO权重作为预训练**，将配置文件中的`pretrain_weights: `设置为对应COCO模型训好的权重，一般会提示head分类层卷积的通道数没对应上，属于正常现象，是由于自定义数据集一般和COCO数据集种类数不一致；\n- YOLO检测模型建议**总`batch_size`至少大于`64`**去训练，如果资源不够请**换小模型**或**减小模型的输入尺度**，为了保障较高检测精度，**尽量不要尝试单卡训和总`batch_size`小于`64`训**；\n\n\u003c/details\u003e\n\n\n## 更新日志\n\n* 【2024/07/10】新增昇腾910B硬件支持，支持的模型见 [模型列表](docs/hardware/supported_models.md)\n* 【2023/05/21】支持[RT-DETR](configs/rtdetr)、[YOLOv8](configs/yolov8)、[YOLOv5u](configs/yolov5/yolov5u)和[YOLOv7u](configs/yolov7/yolov7u)训练全流程，支持[YOLOv6Lite](configs/yolov6/yolov6lite)预测和部署；\n* 【2023/03/13】支持[YOLOv5u](configs/yolov5/yolov5u)和[YOLOv7u](configs/yolov7/yolov7u)预测和部署；\n* 【2023/01/10】支持[YOLOv8](configs/yolov8)预测和部署；\n* 【2022/09/29】支持[RTMDet](configs/rtmdet)预测和部署；\n* 【2022/09/26】发布[PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO)模型套件，请参照[ModelZoo](docs/MODEL_ZOO_cn.md)；\n* 【2022/09/19】支持[YOLOv6](configs/yolov6)新版，包括n/t/s/m/l模型；\n* 【2022/08/23】发布`YOLOSeries`代码库: 支持`YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6`,`YOLOv7`等YOLO模型，支持`ConvNeXt`骨干网络高精度版`PP-YOLOE`,`YOLOX`和`YOLOv5`等模型，支持PaddleSlim无损加速量化训练`PP-YOLOE`,`YOLOv5`,`YOLOv6`和`YOLOv7`等模型，详情可阅读[此文章](https://mp.weixin.qq.com/s/Hki01Zs2lQgvLSLWS0btrA)；\n\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157793354-6e7f381a-0aa6-4bb7-845c-9acf2ecc05c3.png\" width=\"20\"/\u003e 产品动态\n\n- 🔥 **2023.3.14：PaddleYOLO发布[release/2.6版本](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.6)**\n  - 💡 模型套件：\n    - 支持`YOLOv8`,`YOLOv5u`,`YOLOv7u`等YOLO模型预测和部署；\n    - 支持`Swin-Transformer`、`ViT`、`FocalNet`骨干网络高精度版`PP-YOLOE+`等模型；\n    - 支持`YOLOv8`在[FastDeploy](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection)中多硬件快速部署；\n\n- 🔥 **2022.9.26：PaddleYOLO发布[release/2.5版本](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.5)**\n  - 💡 模型套件：\n    - 发布[PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO)模型套件: 支持`YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6`,`YOLOv7`等YOLO模型，支持`ConvNeXt`骨干网络高精度版`PP-YOLOE`,`YOLOX`和`YOLOv5`等模型，支持PaddleSlim无损加速量化训练`PP-YOLOE`,`YOLOv5`,`YOLOv6`和`YOLOv7`等模型；\n\n- 🔥 **2022.8.26：PaddleDetection发布[release/2.5版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5)**\n  - 🗳 特色模型：\n    - 发布[PP-YOLOE+](configs/ppyoloe)，最高精度提升2.4% mAP，达到54.9% mAP，模型训练收敛速度提升3.75倍，端到端预测速度最高提升2.3倍；多个下游任务泛化性提升\n    - 发布[PicoDet-NPU](configs/picodet)模型，支持模型全量化部署；新增[PicoDet](configs/picodet)版面分析模型\n    - 发布[PP-TinyPose升级版](./configs/keypoint/tiny_pose/)增强版，在健身、舞蹈等场景精度提升9.1% AP，支持侧身、卧躺、跳跃、高抬腿等非常规动作\n  - 🔮 场景能力：\n    - 发布行人分析工具[PP-Human v2](./deploy/pipeline)，新增打架、打电话、抽烟、闯入四大行为识别，底层算法性能升级，覆盖行人检测、跟踪、属性三类核心算法能力，提供保姆级全流程开发及模型优化策略，支持在线视频流输入\n    - 首次发布[PP-Vehicle](./deploy/pipeline)，提供车牌识别、车辆属性分析（颜色、车型）、车流量统计以及违章检测四大功能，兼容图片、在线视频流、视频输入，提供完善的二次开发文档教程\n  - 💡 前沿算法：\n    - 全面覆盖的[YOLO家族](https://github.com/PaddlePaddle/PaddleYOLO)经典与最新模型: 包括YOLOv3，百度飞桨自研的实时高精度目标检测检测模型PP-YOLOE，以及前沿检测算法YOLOv4、YOLOv5、YOLOX，YOLOv6及YOLOv7\n    - 新增基于[ViT](configs/vitdet)骨干网络高精度检测模型，COCO数据集精度达到55.7% mAP；新增[OC-SORT](configs/mot/ocsort)多目标跟踪模型；新增[ConvNeXt](configs/convnext)骨干网络\n  - 📋 产业范例：新增[智能健身](https://aistudio.baidu.com/aistudio/projectdetail/4385813)、[打架识别](https://aistudio.baidu.com/aistudio/projectdetail/4086987?channelType=0\u0026channel=0)、[来客分析](https://aistudio.baidu.com/aistudio/projectdetail/4230123?channelType=0\u0026channel=0)、车辆结构化范例\n\n- 2022.3.24：PaddleDetection发布[release/2.4版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4)\n  - 发布高精度云边一体SOTA目标检测模型[PP-YOLOE](configs/ppyoloe)，提供s/m/l/x版本，l版本COCO test2017数据集精度51.6%，V100预测速度78.1 FPS，支持混合精度训练，训练较PP-YOLOv2加速33%，全系列多尺度模型，满足不同硬件算力需求，可适配服务器、边缘端GPU及其他服务器端AI加速卡。\n  - 发布边缘端和CPU端超轻量SOTA目标检测模型[PP-PicoDet增强版](configs/picodet)，精度提升2%左右，CPU预测速度提升63%，新增参数量0.7M的PicoDet-XS模型，提供模型稀疏化和量化功能，便于模型加速，各类硬件无需单独开发后处理模块，降低部署门槛。\n  - 发布实时行人分析工具[PP-Human](deploy/pipeline)，支持行人跟踪、人流量统计、人体属性识别与摔倒检测四大能力，基于真实场景数据特殊优化，精准识别各类摔倒姿势，适应不同环境背景、光线及摄像角度。\n  - 新增[YOLOX](configs/yolox)目标检测模型，支持nano/tiny/s/m/l/x版本，x版本COCO val2017数据集精度51.8%。\n\n- [更多版本发布](https://github.com/PaddlePaddle/PaddleDetection/releases)\n\n## \u003cimg title=\"\" src=\"https://user-images.githubusercontent.com/48054808/157795569-9fc77c85-732f-4870-9be0-99a7fe2cff27.png\" alt=\"\" width=\"20\"\u003e 简介\n\n**PaddleDetection**为基于飞桨PaddlePaddle的端到端目标检测套件，内置**30+模型算法**及**250+预训练模型**，覆盖**目标检测、实例分割、跟踪、关键点检测**等方向，其中包括**服务器端和移动端高精度、轻量级**产业级SOTA模型、冠军方案和学术前沿算法，并提供配置化的网络模块组件、十余种数据增强策略和损失函数等高阶优化支持和多种部署方案，在打通数据处理、模型开发、训练、压缩、部署全流程的基础上，提供丰富的案例及教程，加速算法产业落地应用。\n\n\u003cdiv  align=\"center\"\u003e\n  \u003cimg src=\"https://user-images.githubusercontent.com/22989727/189026616-75f9c06c-b403-4a61-9372-0fcbed6e0662.gif\" width=\"800\"/\u003e\n\u003c/div\u003e\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157799599-e6a66855-bac6-4e75-b9c0-96e13cb9612f.png\" width=\"20\"/\u003e 特性\n\n- **模型丰富**: 包含**目标检测**、**实例分割**、**人脸检测**、****关键点检测****、**多目标跟踪**等**250+个预训练模型**，涵盖多种**全球竞赛冠军**方案。\n- **使用简洁**：模块化设计，解耦各个网络组件，开发者轻松搭建、试用各种检测模型及优化策略，快速得到高性能、定制化的算法。\n- **端到端打通**: 从数据增强、组网、训练、压缩、部署端到端打通，并完备支持**云端**/**边缘端**多架构、多设备部署。\n- **高性能**: 基于飞桨的高性能内核，模型训练速度及显存占用优势明显。支持FP16训练, 支持多机训练。\n\n\u003cdiv  align=\"center\"\u003e\n  \u003cimg src=\"https://user-images.githubusercontent.com/22989727/189026189-5d21e93a-5b33-40ce-bc36-c737122c1992.png\" width=\"800\"/\u003e\n\u003c/div\u003e\n\n## \u003cimg title=\"\" src=\"https://user-images.githubusercontent.com/48054808/157800467-2a9946ad-30d1-49a9-b9db-ba33413d9c90.png\" alt=\"\" width=\"20\"\u003e 技术交流\n\n- 如果你发现任何PaddleDetection存在的问题或者是建议, 欢迎通过[GitHub Issues](https://github.com/PaddlePaddle/PaddleDetection/issues)给我们提issues。\n\n- **欢迎加入PaddleDetection 微信用户群（扫码填写问卷即可入群）**\n  - **入群福利 💎：获取PaddleDetection团队整理的重磅学习大礼包🎁**\n    - 📊 福利一：获取飞桨联合业界企业整理的开源数据集\n    - 👨‍🏫 福利二：获取PaddleDetection历次发版直播视频与最新直播咨询\n    - 🗳 福利三：获取垂类场景预训练模型集合，包括工业、安防、交通等5+行业场景\n    - 🗂 福利四：获取10+全流程产业实操范例，覆盖火灾烟雾检测、人流量计数等产业高频场景\n  \u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"https://user-images.githubusercontent.com/34162360/177678712-4655747d-4290-4ad9-b7a1-4564a5418ac6.jpg\"  width = \"200\" /\u003e  \n  \u003c/div\u003e\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157827140-03ffaff7-7d14-48b4-9440-c38986ea378c.png\" width=\"20\"/\u003e 套件结构概览\n\n\u003ctable align=\"center\"\u003e\n  \u003ctbody\u003e\n    \u003ctr align=\"center\" valign=\"bottom\"\u003e\n      \u003ctd\u003e\n        \u003cb\u003eArchitectures\u003c/b\u003e\n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cb\u003eBackbones\u003c/b\u003e\n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cb\u003eComponents\u003c/b\u003e\n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cb\u003eData Augmentation\u003c/b\u003e\n      \u003c/td\u003e\n    \u003c/tr\u003e\n    \u003ctr valign=\"top\"\u003e\n      \u003ctd\u003e\n        \u003cul\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eObject Detection\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eYOLOv3\u003c/li\u003e  \n            \u003cli\u003eYOLOv5\u003c/li\u003e  \n            \u003cli\u003eYOLOv6\u003c/li\u003e  \n            \u003cli\u003eYOLOv7\u003c/li\u003e  \n            \u003cli\u003eYOLOv8\u003c/li\u003e  \n            \u003cli\u003ePP-YOLOv1/v2\u003c/li\u003e\n            \u003cli\u003ePP-YOLO-Tiny\u003c/li\u003e\n            \u003cli\u003ePP-YOLOE\u003c/li\u003e\n            \u003cli\u003ePP-YOLOE+\u003c/li\u003e\n            \u003cli\u003eYOLOX\u003c/li\u003e\n            \u003cli\u003eRTMDet\u003c/li\u003e\n         \u003c/ul\u003e\u003c/details\u003e\n      \u003c/ul\u003e\n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eDetails\u003c/b\u003e\u003c/summary\u003e\n        \u003cul\u003e\n          \u003cli\u003eResNet(\u0026vd)\u003c/li\u003e\n          \u003cli\u003eCSPResNet\u003c/li\u003e\n          \u003cli\u003eDarkNet\u003c/li\u003e\n          \u003cli\u003eCSPDarkNet\u003c/li\u003e\n          \u003cli\u003eConvNeXt\u003c/li\u003e\n          \u003cli\u003eEfficientRep\u003c/li\u003e\n          \u003cli\u003eCSPBepBackbone\u003c/li\u003e\n          \u003cli\u003eELANNet\u003c/li\u003e\n          \u003cli\u003eCSPNeXt\u003c/li\u003e\n        \u003c/ul\u003e\u003c/details\u003e\n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eCommon\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eSync-BN\u003c/li\u003e\n            \u003cli\u003eGroup Norm\u003c/li\u003e\n            \u003cli\u003eDCNv2\u003c/li\u003e\n            \u003cli\u003eEMA\u003c/li\u003e\n          \u003c/ul\u003e \u003c/details\u003e\n        \u003c/ul\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eFPN\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eYOLOv3FPN\u003c/li\u003e\n            \u003cli\u003ePPYOLOFPN\u003c/li\u003e\n            \u003cli\u003ePPYOLOTinyFPN\u003c/li\u003e\n            \u003cli\u003ePPYOLOPAN\u003c/li\u003e\n            \u003cli\u003eYOLOCSPPAN\u003c/li\u003e\n            \u003cli\u003eCustom-PAN\u003c/li\u003e\n            \u003cli\u003eRepPAN\u003c/li\u003e\n            \u003cli\u003eCSPRepPAN\u003c/li\u003e\n            \u003cli\u003eELANFPN\u003c/li\u003e\n            \u003cli\u003eELANFPNP6\u003c/li\u003e\n            \u003cli\u003eCSPNeXtPAFPN\u003c/li\u003e\n          \u003c/ul\u003e \u003c/details\u003e\n        \u003c/ul\u003e  \n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eLoss\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eSmooth-L1\u003c/li\u003e\n            \u003cli\u003eGIoU/DIoU/CIoU\u003c/li\u003e  \n            \u003cli\u003eIoUAware\u003c/li\u003e\n            \u003cli\u003eFocal Loss\u003c/li\u003e\n            \u003cli\u003eVariFocal Loss\u003c/li\u003e\n          \u003c/ul\u003e \u003c/details\u003e\n        \u003c/ul\u003e  \n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003ePost-processing\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eSoftNMS\u003c/li\u003e\n            \u003cli\u003eMatrixNMS\u003c/li\u003e  \n          \u003c/ul\u003e \u003c/details\u003e  \n        \u003c/ul\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eSpeed\u003c/b\u003e\u003c/summary\u003e\n          \u003cul\u003e\n            \u003cli\u003eFP16 training\u003c/li\u003e\n            \u003cli\u003eMulti-machine training \u003c/li\u003e  \n          \u003c/ul\u003e \u003c/details\u003e  \n        \u003c/ul\u003e  \n      \u003c/td\u003e\n      \u003ctd\u003e\n        \u003cdetails open\u003e\u003csummary\u003e\u003cb\u003eDetails\u003c/b\u003e\u003c/summary\u003e\n        \u003cul\u003e\n          \u003cli\u003eResize\u003c/li\u003e  \n          \u003cli\u003eLighting\u003c/li\u003e  \n          \u003cli\u003eFlipping\u003c/li\u003e  \n          \u003cli\u003eExpand\u003c/li\u003e\n          \u003cli\u003eCrop\u003c/li\u003e\n          \u003cli\u003eColor Distort\u003c/li\u003e  \n          \u003cli\u003eRandom Erasing\u003c/li\u003e  \n          \u003cli\u003eMixup \u003c/li\u003e\n          \u003cli\u003eAugmentHSV\u003c/li\u003e\n          \u003cli\u003eMosaic\u003c/li\u003e\n          \u003cli\u003eCutmix \u003c/li\u003e\n          \u003cli\u003eGrid Mask\u003c/li\u003e\n          \u003cli\u003eAuto Augment\u003c/li\u003e  \n          \u003cli\u003eRandom Perspective\u003c/li\u003e  \n        \u003c/ul\u003e \u003c/details\u003e  \n      \u003c/td\u003e  \n    \u003c/tr\u003e\n\n\u003c/td\u003e\n    \u003c/tr\u003e\n  \u003c/tbody\u003e\n\u003c/table\u003e\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157801371-9a9a8c65-1690-4123-985a-e0559a7f9494.png\" width=\"20\"/\u003e 模型性能概览\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cb\u003e 云端模型性能对比\u003c/b\u003e\u003c/summary\u003e\n\n各模型结构和骨干网络的代表模型在COCO数据集上精度mAP和单卡Tesla V100上预测速度(FPS)对比图。\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"docs/images/fps_map.png\" /\u003e\n\u003c/div\u003e\n\n**说明：**\n\n- `PP-YOLOE`是对`PP-YOLO v2`模型的进一步优化，在COCO数据集精度51.6%，Tesla V100预测速度78.1FPS\n- `PP-YOLOE+`是对`PPOLOE`模型的进一步优化，在COCO数据集精度53.3%，Tesla V100预测速度78.1FPS\n- 图中模型均可在[模型库](#模型库)中获取\n\n\u003c/details\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cb\u003e 移动端模型性能对比\u003c/b\u003e\u003c/summary\u003e\n\n各移动端模型在COCO数据集上精度mAP和高通骁龙865处理器上预测速度(FPS)对比图。\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"docs/images/mobile_fps_map.png\" width=600/\u003e\n\u003c/div\u003e\n\n**说明：**\n\n- 测试数据均使用高通骁龙865(4\\*A77 + 4\\*A55)处理器batch size为1, 开启4线程测试，测试使用NCNN预测库，测试脚本见[MobileDetBenchmark](https://github.com/JiweiMaster/MobileDetBenchmark)\n- [PP-PicoDet](configs/picodet)及[PP-YOLO-Tiny](configs/ppyolo)为PaddleDetection自研模型，其余模型PaddleDetection暂未提供\n\n\u003c/details\u003e\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157829890-a535b8a6-631c-4c87-b861-64d4b32b2d6a.png\" width=\"20\"/\u003e 模型库\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cb\u003e 1. 通用检测\u003c/b\u003e\u003c/summary\u003e\n\n#### [PP-YOLOE+](./configs/ppyoloe)系列 推荐场景：Nvidia V100, T4等云端GPU和Jetson系列等边缘端设备\n\n| 模型名称       | COCO精度（mAP） | V100 TensorRT FP16速度(FPS) | 配置文件                                                  | 模型下载                                                                                 |\n|:---------- |:-----------:|:-------------------------:|:-----------------------------------------------------:|:------------------------------------------------------------------------------------:|\n| PP-YOLOE+_s | 43.9        | 333.3                     | [链接](configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml)     | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_s_80e_coco.pdparams)      |\n| PP-YOLOE+_m | 50.0        | 208.3                     | [链接](configs/ppyoloe/ppyoloe_plus_crn_m_80e_coco.yml)     | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams)     |\n| PP-YOLOE+_l | 53.3        | 149.2                     | [链接](configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams) |\n| PP-YOLOE+_x | 54.9        | 95.2                      | [链接](configs/ppyoloe/ppyoloe_plus_crn_x_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_x_80e_coco.pdparams) |\n\n#### 前沿检测算法\n\n| 模型名称                                                               | COCO精度（mAP） | V100 TensorRT FP16速度(FPS) | 配置文件                                                                                                         | 模型下载                                                                       |\n|:------------------------------------------------------------------ |:-----------:|:-------------------------:|:------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------:|\n| [YOLOX-l](configs/yolox)                                           | 50.1        | 107.5                     | [链接](configs/yolox/yolox_l_300e_coco.yml)                                                                    | [下载地址](https://paddledet.bj.bcebos.com/models/yolox_l_300e_coco.pdparams)  |\n| [YOLOv5-l](configs/yolov5) | 48.6        | 136.0                     | [链接](configs/yolov5/yolov5_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolov5_l_300e_coco.pdparams) |\n| [YOLOv7-l](configs/yolov7) | 51.0        | 135.0                     | [链接](configs/yolov7/yolov7_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolov7_l_300e_coco.pdparams) |\n\n\u003c/details\u003e\n\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157828296-d5eb0ccb-23ea-40f5-9957-29853d7d13a9.png\" width=\"20\"/\u003e 文档教程\n\n### 入门教程\n\n- [安装说明](docs/tutorials/INSTALL_cn.md)\n- [快速体验](docs/tutorials/QUICK_STARTED_cn.md)\n- [数据准备](docs/tutorials/data/README.md)\n- [PaddleDetection全流程使用](docs/tutorials/GETTING_STARTED_cn.md)\n- [FAQ/常见问题汇总](docs/tutorials/FAQ)\n\n### 进阶教程\n\n- 参数配置\n\n  - [PP-YOLO参数说明](docs/tutorials/config_annotation/ppyolo_r50vd_dcn_1x_coco_annotation.md)\n\n- 模型压缩(基于[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim))\n\n  - [剪裁/量化/蒸馏教程](configs/slim)\n\n- [推理部署](deploy/README.md)\n\n  - [模型导出教程](deploy/EXPORT_MODEL.md)\n  - [Paddle Inference部署](deploy/README.md)\n    - [Python端推理部署](deploy/python)\n    - [C++端推理部署](deploy/cpp)\n  - [Paddle-Lite部署](deploy/lite)\n  - [Paddle Serving部署](deploy/serving)\n  - [ONNX模型导出](deploy/EXPORT_ONNX_MODEL.md)\n  - [推理benchmark](deploy/BENCHMARK_INFER.md)\n\n- 进阶开发\n\n  - [数据处理模块](docs/advanced_tutorials/READER.md)\n  - [新增检测模型](docs/advanced_tutorials/MODEL_TECHNICAL.md)\n  - 二次开发教程\n    - [目标检测](docs/advanced_tutorials/customization/detection.md)\n\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157835981-ef6057b4-6347-4768-8fcc-cd07fcc3d8b0.png\" width=\"20\"/\u003e 版本更新\n\n版本更新内容请参考[版本更新文档](docs/CHANGELOG.md)\n\n\n## \u003cimg title=\"\" src=\"https://user-images.githubusercontent.com/48054808/157835345-f5d24128-abaf-4813-b793-d2e5bdc70e5a.png\" alt=\"\" width=\"20\"\u003e 许可证书\n\n本项目的发布受[GPL-3.0 license](LICENSE)许可认证。\n\n\n## \u003cimg src=\"https://user-images.githubusercontent.com/48054808/157835276-9aab9d1c-1c46-446b-bdd4-5ab75c5cfa48.png\" width=\"20\"/\u003e 引用\n\n```\n@misc{ppdet2019,\ntitle={PaddleDetection, Object detection and instance segmentation toolkit based on PaddlePaddle.},\nauthor={PaddlePaddle Authors},\nhowpublished = {\\url{https://github.com/PaddlePaddle/PaddleDetection}},\nyear={2019}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FPaddlePaddle%2FPaddleYOLO","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FPaddlePaddle%2FPaddleYOLO","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FPaddlePaddle%2FPaddleYOLO/lists"}