{"id":13641615,"url":"https://github.com/UNeedCryDear/yolov5-seg-opencv-onnxruntime-cpp","last_synced_at":"2025-04-20T11:31:26.028Z","repository":{"id":61727789,"uuid":"548743725","full_name":"UNeedCryDear/yolov5-seg-opencv-onnxruntime-cpp","owner":"UNeedCryDear","description":"yolov5 segmentation with onnxruntime and opencv","archived":false,"fork":false,"pushed_at":"2024-04-10T11:10:24.000Z","size":4270,"stargazers_count":149,"open_issues_count":4,"forks_count":30,"subscribers_count":4,"default_branch":"master","last_synced_at":"2024-08-03T01:24:05.010Z","etag":null,"topics":["onnxruntime","opencv","yolov5-seg"],"latest_commit_sha":null,"homepage":"","language":"C++","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/UNeedCryDear.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":"2022-10-10T05:40:29.000Z","updated_at":"2024-07-31T13:09:49.000Z","dependencies_parsed_at":"2023-02-09T02:31:30.075Z","dependency_job_id":"e41f86fa-f64c-4ea4-a854-98c0af9b7951","html_url":"https://github.com/UNeedCryDear/yolov5-seg-opencv-onnxruntime-cpp","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/UNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/UNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/UNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/UNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/UNeedCryDear","download_url":"https://codeload.github.com/UNeedCryDear/yolov5-seg-opencv-onnxruntime-cpp/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":223827479,"owners_count":17209796,"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":["onnxruntime","opencv","yolov5-seg"],"created_at":"2024-08-02T01:01:22.300Z","updated_at":"2024-11-09T12:30:34.402Z","avatar_url":"https://github.com/UNeedCryDear.png","language":"C++","funding_links":[],"categories":["Other Versions of YOLO"],"sub_categories":[],"readme":"# yolov5-seg-opencv-onnxruntime-cpp\n使用opencv-dnn部署yolov5实例分割模型\n基于6.2版本的yolov5:https://github.com/ultralytics/yolov5\n\n**OpenCV\u003e=4.5.0**\u003cbr\u003e\n**ONNXRuntime\u003e=1.9.0,** (Maybe the earlier version of onnxruntime is also possible, but I didn't test it)\n```\n# for oprncv\npython export.py --weights yolov5s-seg.pt --img [640,640] --include onnx --opset 12\n# for onnxruntime\npython export.py --weights yolov5s-seg.pt --img [640,640] --include onnx  #static\npython export.py --weights yolov5s-seg.pt  --batch-size bs-number --dynamic --include onnx  #dyamic\n```\n#### 2023.11.09更新\u003cbr\u003e\n+ 将yolov5的opencv推理合并进来，并且由于之前的其他作者的onnxruntime的部署由于ORT的版本更新导致出现问题，所以我这词更新新增onnxruntime推理，以适应新版本的ORT。\n+ 修复此pr中提到的一些问题[https://github.com/UNeedCryDear/yolov8-opencv-onnxruntime-cpp/pull/30]，此bug会导致mask与box大小可能会差几个像素从而导致出现一些问题（如果用的时候没有注意的话），本次更新之后会将其缩放到一致大小。\n+ 新增视频流推理的demo，这是由于发现很多初学者调用视频的时候总是每一张图片都去读取一次模型，所以本次更新一起加上去。\n\n\n#### 2023.09.20更新\u003cbr\u003e\n+ 新增模型路径检查，部分issue查了半天，发现模型路径不对。\n+ 计算mask部分bug修复，此前如果输入大小非640的话，需要同时设置头文件和结构体才能完成检测，但是大部分人只修改了一个地方，目前优化这部分内容，只需要修头文件中的定义即可。另外将segHeight和segWidth设置为从网络输出中读取，这样如果mask-ratio不是4倍的话，可以不需要修改这两个参数值。\n+ 修复```GetMask2()```中可能导致越界的问题。\u003cbr\u003e\n#### 2023.01.11 更新：\n+ 目前opencv4.7.0的版本会有问题（https://github.com/opencv/opencv/issues/23080) ，如果你是opencv4.7.0的版本，你需要在```net.forward()``` 前面加上```net.enableWinograd(false);```来关闭Winograd加速。\n\n\n#### 2022.12.19 更新：\n+ **new:** 新增加onnxruntime推理，支持onnx的动态推理和批量推理，避免opencv不支持动态的尴尬境地。\n+ onnxruntime的版本最低要求目前未知，我仅仅测试了ort12.0+ort13.0这两个大版本(11.0的应该问题也不大),如果有人测试比这些更低的版本可以运行通过，可以通知我一下。\n+ 为了兼容，代码结构有部分变动。\n#### 2022.12.13 更新：\n+ 如果你的显卡支持FP16推理的话，可以将模型读取代码中的```DNN_TARGET_CUDA```改成```DNN_TARGET_CUDA_FP16```提升推理速度（虽然是蚊子腿，好歹也是肉（： \n#### 2022.11.10 更新：\n+ 之前旧版本计算结果mask图像的时候，是整个特征图和特征掩码进行的矩乘法，然后切割出bounding-box区域二值化，这个速度在我看来还是太慢，特别是检测结果框多并且目标都不大的情况下，整张特征图进行乘法的开销大，而有用的就区域就一小块，类似下面的领带那样，大部分特征图区域是无效的，所以本次修改变成对特征图进行裁剪，然后进行后续的矩阵乘法，特别是小目标比较多的情况下提升比较大，而两者结果基本上基本一致，仅仅由于部分四舍五入的原因差了一个像素。如果你的小目标对于一个像素的偏移无法接受的话，那么就使用旧版本方法。\n+ 这两种方法具体差距情况可以看[differ-from-tow-method.bmp](res/bus_diff.bmp)。\n\n \n#### 2022.10.10 更新：  \n+ 0.opencv不支持动态推理，请不要加--dymanic导出onnx。\n+ 1.关于换行符，windows下面需要设置为CRLF，上传到github会自动切换成LF，windows下面切换一下即可\u003cbr\u003e\n+ 2.有些小伙伴用版本为1.12.x的pytorch的时候，需要将\nhttps://github.com/ultralytics/yolov5/blob/c98128fe71a8676037a0605ab389c7473c743d07/export.py#L155\n这里的标志位改成```do_constant_folding=False, ```,否者opencv用dnn读取不了onnx文件\n+ **4.关于mask**\n  \u003e 原始的mask采用的是整张图片mask，即使你的box很小，整张图就一小块区域，也会是整个mask。修改之后变成了box内mask，mask跟着box的大小走（原始代码中的crop操作），提升速度的同时，内存开销也会减小。具体变换结果可以看[tie-mask.bmp](res/boxMask.bmp)\n\n以下为yolov5-seg.onnx运行结果：\n![](res/bus.bmp)\n![](res/zidane.bmp)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FUNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FUNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FUNeedCryDear%2Fyolov5-seg-opencv-onnxruntime-cpp/lists"}