[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-71963":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":27,"readmeContent":28,"aiSummary":29,"trendingCount":16,"starSnapshotCount":16,"syncStatus":30,"lastSyncTime":31,"discoverSource":32},71963,"yolov9","WongKinYiu\u002Fyolov9","WongKinYiu","Implementation of paper - YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information","",null,"Python",9530,1621,56,384,0,3,4,15,9,75.13,"GNU General Public License v3.0",false,"main",true,[5],"2026-06-12 04:01:03","# YOLOv9\n\nImplementation of paper - [YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.13616)\n\n[![arxiv.org](http:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcs.CV-arXiv%3A2402.13616-B31B1B.svg)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.13616)\n[![Hugging Face Spaces](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fkadirnar\u002FYolov9)\n[![Hugging Face Spaces](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https:\u002F\u002Fhuggingface.co\u002Fmerve\u002Fyolov9)\n[![Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Froboflow-ai\u002Fnotebooks\u002Fblob\u002Fmain\u002Fnotebooks\u002Ftrain-yolov9-object-detection-on-custom-dataset.ipynb)\n[![OpenCV](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FOpenCV-BlogPost-black?logo=opencv&labelColor=blue&color=black)](https:\u002F\u002Flearnopencv.com\u002Fyolov9-advancing-the-yolo-legacy\u002F)\n\n\u003Cdiv align=\"center\">\n    \u003Ca href=\".\u002F\">\n        \u003Cimg src=\".\u002Ffigure\u002Fperformance.png\" width=\"79%\"\u002F>\n    \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\n## Performance \n\nMS COCO\n\n| Model | Test Size | AP\u003Csup>val\u003C\u002Fsup> | AP\u003Csub>50\u003C\u002Fsub>\u003Csup>val\u003C\u002Fsup> | AP\u003Csub>75\u003C\u002Fsub>\u003Csup>val\u003C\u002Fsup> | Param. | FLOPs |\n| :-- | :-: | :-: | :-: | :-: | :-: | :-: |\n| [**YOLOv9-T**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-t-converted.pt) | 640 | **38.3%** | **53.1%** | **41.3%** | **2.0M** | **7.7G** |\n| [**YOLOv9-S**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-s-converted.pt) | 640 | **46.8%** | **63.4%** | **50.7%** | **7.1M** | **26.4G** |\n| [**YOLOv9-M**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-m-converted.pt) | 640 | **51.4%** | **68.1%** | **56.1%** | **20.0M** | **76.3G** |\n| [**YOLOv9-C**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-c-converted.pt) | 640 | **53.0%** | **70.2%** | **57.8%** | **25.3M** | **102.1G** |\n| [**YOLOv9-E**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-e-converted.pt) | 640 | **55.6%** | **72.8%** | **60.6%** | **57.3M** | **189.0G** |\n\u003C!-- | [**YOLOv9 (ReLU)**]() | 640 | **51.9%** | **69.1%** | **56.5%** | **25.3M** | **102.1G** | -->\n\n\u003C!-- tiny, small, and medium models will be released after the paper be accepted and published. -->\n\n## Useful Links\n\n\u003Cdetails>\u003Csummary> \u003Cb>Expand\u003C\u002Fb> \u003C\u002Fsummary>\n\nCustom training: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F30#issuecomment-1960955297\n    \nONNX export: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F2#issuecomment-1960519506 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F40#issue-2150697688 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F130#issue-2162045461\n\nONNX export for segmentation: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F260#issue-2191162150\n\nTensorRT inference: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F143#issuecomment-1975049660 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F34#issue-2150393690 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F79#issue-2153547004 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F143#issue-2164002309\n\nQAT TensorRT: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F327#issue-2229284136 https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F253#issue-2189520073\n\nTensorRT inference for segmentation: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F446\n\nTFLite: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F374#issuecomment-2065751706\n\nOpenVINO: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F164#issue-2168540003\n\nC# ONNX inference: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F95#issue-2155974619\n\nC# OpenVINO inference: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F95#issuecomment-1968131244\n\nOpenCV: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F113#issuecomment-1971327672\n\nHugging Face demo: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F45#issuecomment-1961496943\n\nCoLab demo: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fpull\u002F18\n\nONNXSlim export: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fpull\u002F37\n\nYOLOv9 ROS: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F144#issue-2164210644\n\nYOLOv9 ROS TensorRT: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F145#issue-2164218595\n\nYOLOv9 Julia: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F141#issuecomment-1973710107\n\nYOLOv9 MLX: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F258#issue-2190586540\n\nYOLOv9 StrongSORT with OSNet: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F299#issue-2212093340\n\nYOLOv9 ByteTrack: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F78#issue-2153512879\n\nYOLOv9 DeepSORT: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F98#issue-2156172319\n\nYOLOv9 counting: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F84#issue-2153904804\n\nYOLOv9 speed estimation: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F456\n\nYOLOv9 face detection: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F121#issue-2160218766\n\nYOLOv9 segmentation onnxruntime: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F151#issue-2165667350\n\nComet logging: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fpull\u002F110\n\nMLflow logging: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fpull\u002F87\n\nAnyLabeling tool: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F48#issue-2152139662\n\nAX650N deploy: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F96#issue-2156115760\n\nConda environment: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fpull\u002F93\n\nAutoDL docker environment: https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fissues\u002F112#issue-2158203480\n\n\u003C\u002Fdetails>\n\n\n## Installation\n\nDocker environment (recommended)\n\u003Cdetails>\u003Csummary> \u003Cb>Expand\u003C\u002Fb> \u003C\u002Fsummary>\n\n``` shell\n# create the docker container, you can change the share memory size if you have more.\nnvidia-docker run --name yolov9 -it -v your_coco_path\u002F:\u002Fcoco\u002F -v your_code_path\u002F:\u002Fyolov9 --shm-size=64g nvcr.io\u002Fnvidia\u002Fpytorch:21.11-py3\n\n# apt install required packages\napt update\napt install -y zip htop screen libgl1-mesa-glx\n\n# pip install required packages\npip install seaborn thop\n\n# go to code folder\ncd \u002Fyolov9\n```\n\n\u003C\u002Fdetails>\n\n\n## Evaluation\n\n[`yolov9-s-converted.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-s-converted.pt) [`yolov9-m-converted.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-m-converted.pt) [`yolov9-c-converted.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-c-converted.pt) [`yolov9-e-converted.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-e-converted.pt) \n[`yolov9-s.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-s.pt) [`yolov9-m.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-m.pt) [`yolov9-c.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-c.pt) [`yolov9-e.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fyolov9-e.pt) \n[`gelan-s.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-s.pt) [`gelan-m.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-m.pt) [`gelan-c.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c.pt) [`gelan-e.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-e.pt)\n\n``` shell\n# evaluate converted yolov9 models\npython val.py --data data\u002Fcoco.yaml --img 640 --batch 32 --conf 0.001 --iou 0.7 --device 0 --weights '.\u002Fyolov9-c-converted.pt' --save-json --name yolov9_c_c_640_val\n\n# evaluate yolov9 models\n# python val_dual.py --data data\u002Fcoco.yaml --img 640 --batch 32 --conf 0.001 --iou 0.7 --device 0 --weights '.\u002Fyolov9-c.pt' --save-json --name yolov9_c_640_val\n\n# evaluate gelan models\n# python val.py --data data\u002Fcoco.yaml --img 640 --batch 32 --conf 0.001 --iou 0.7 --device 0 --weights '.\u002Fgelan-c.pt' --save-json --name gelan_c_640_val\n```\n\nYou will get the results:\n\n```\n Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.530\n Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.702\n Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.578\n Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.362\n Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.585\n Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.693\n Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.392\n Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.652\n Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.702\n Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.541\n Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.760\n Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.844\n```\n\n\n## Training\n\nData preparation\n\n``` shell\nbash scripts\u002Fget_coco.sh\n```\n\n* Download MS COCO dataset images ([train](http:\u002F\u002Fimages.cocodataset.org\u002Fzips\u002Ftrain2017.zip), [val](http:\u002F\u002Fimages.cocodataset.org\u002Fzips\u002Fval2017.zip), [test](http:\u002F\u002Fimages.cocodataset.org\u002Fzips\u002Ftest2017.zip)) and [labels](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov7\u002Freleases\u002Fdownload\u002Fv0.1\u002Fcoco2017labels-segments.zip). If you have previously used a different version of YOLO, we strongly recommend that you delete `train2017.cache` and `val2017.cache` files, and redownload [labels](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov7\u002Freleases\u002Fdownload\u002Fv0.1\u002Fcoco2017labels-segments.zip) \n\nSingle GPU training\n\n``` shell\n# train yolov9 models\npython train_dual.py --workers 8 --device 0 --batch 16 --data data\u002Fcoco.yaml --img 640 --cfg models\u002Fdetect\u002Fyolov9-c.yaml --weights '' --name yolov9-c --hyp hyp.scratch-high.yaml --min-items 0 --epochs 500 --close-mosaic 15\n\n# train gelan models\n# python train.py --workers 8 --device 0 --batch 32 --data data\u002Fcoco.yaml --img 640 --cfg models\u002Fdetect\u002Fgelan-c.yaml --weights '' --name gelan-c --hyp hyp.scratch-high.yaml --min-items 0 --epochs 500 --close-mosaic 15\n```\n\nMultiple GPU training\n\n``` shell\n# train yolov9 models\npython -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train_dual.py --workers 8 --device 0,1,2,3,4,5,6,7 --sync-bn --batch 128 --data data\u002Fcoco.yaml --img 640 --cfg models\u002Fdetect\u002Fyolov9-c.yaml --weights '' --name yolov9-c --hyp hyp.scratch-high.yaml --min-items 0 --epochs 500 --close-mosaic 15\n\n# train gelan models\n# python -m torch.distributed.launch --nproc_per_node 4 --master_port 9527 train.py --workers 8 --device 0,1,2,3 --sync-bn --batch 128 --data data\u002Fcoco.yaml --img 640 --cfg models\u002Fdetect\u002Fgelan-c.yaml --weights '' --name gelan-c --hyp hyp.scratch-high.yaml --min-items 0 --epochs 500 --close-mosaic 15\n```\n\n\n## Re-parameterization\n\nSee [reparameterization.ipynb](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Fblob\u002Fmain\u002Ftools\u002Freparameterization.ipynb).\n\n\n## Inference\n\n\u003Cdiv align=\"center\">\n    \u003Ca href=\".\u002F\">\n        \u003Cimg src=\".\u002Ffigure\u002Fhorses_prediction.jpg\" width=\"49%\"\u002F>\n    \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n``` shell\n# inference converted yolov9 models\npython detect.py --source '.\u002Fdata\u002Fimages\u002Fhorses.jpg' --img 640 --device 0 --weights '.\u002Fyolov9-c-converted.pt' --name yolov9_c_c_640_detect\n\n# inference yolov9 models\n# python detect_dual.py --source '.\u002Fdata\u002Fimages\u002Fhorses.jpg' --img 640 --device 0 --weights '.\u002Fyolov9-c.pt' --name yolov9_c_640_detect\n\n# inference gelan models\n# python detect.py --source '.\u002Fdata\u002Fimages\u002Fhorses.jpg' --img 640 --device 0 --weights '.\u002Fgelan-c.pt' --name gelan_c_c_640_detect\n```\n\n\n## Citation\n\n```\n@article{wang2024yolov9,\n  title={{YOLOv9}: Learning What You Want to Learn Using Programmable Gradient Information},\n  author={Wang, Chien-Yao  and Liao, Hong-Yuan Mark},\n  booktitle={arXiv preprint arXiv:2402.13616},\n  year={2024}\n}\n```\n\n```\n@article{chang2023yolor,\n  title={{YOLOR}-Based Multi-Task Learning},\n  author={Chang, Hung-Shuo and Wang, Chien-Yao and Wang, Richard Robert and Chou, Gene and Liao, Hong-Yuan Mark},\n  journal={arXiv preprint arXiv:2309.16921},\n  year={2023}\n}\n```\n\n\n## Teaser\n\nParts of code of [YOLOR-Based Multi-Task Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.16921) are released in the repository.\n\n\u003Cdiv align=\"center\">\n    \u003Ca href=\".\u002F\">\n        \u003Cimg src=\".\u002Ffigure\u002Fmultitask.png\" width=\"99%\"\u002F>\n    \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n#### Object Detection\n\n[`gelan-c-det.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-det.pt)\n\n`object detection`\n\n``` shell\n# coco\u002Flabels\u002F{split}\u002F*.txt\n# bbox or polygon (1 instance 1 line)\npython train.py --workers 8 --device 0 --batch 32 --data data\u002Fcoco.yaml --img 640 --cfg models\u002Fdetect\u002Fgelan-c.yaml --weights '' --name gelan-c-det --hyp hyp.scratch-high.yaml --min-items 0 --epochs 300 --close-mosaic 10\n```\n\n| Model | Test Size | Param. | FLOPs | AP\u003Csup>box\u003C\u002Fsup> |\n| :-- | :-: | :-: | :-: | :-: |\n| [**GELAN-C-DET**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-det.pt) | 640 | 25.3M | 102.1G |**52.3%** |\n| [**YOLOv9-C-DET**]() | 640 | 25.3M | 102.1G | **53.0%** |\n\n#### Instance Segmentation\n\n[`gelan-c-seg.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-seg.pt)\n\n`object detection` `instance segmentation`\n\n``` shell\n# coco\u002Flabels\u002F{split}\u002F*.txt\n# polygon (1 instance 1 line)\npython segment\u002Ftrain.py --workers 8 --device 0 --batch 32  --data coco.yaml --img 640 --cfg models\u002Fsegment\u002Fgelan-c-seg.yaml --weights '' --name gelan-c-seg --hyp hyp.scratch-high.yaml --no-overlap --epochs 300 --close-mosaic 10\n```\n\n| Model | Test Size | Param. | FLOPs | AP\u003Csup>box\u003C\u002Fsup> | AP\u003Csup>mask\u003C\u002Fsup>  |\n| :-- | :-: | :-: | :-: | :-: | :-: |\n| [**GELAN-C-SEG**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-seg.pt) | 640 | 27.4M | 144.6G | **52.3%** | **42.4%** |\n| [**YOLOv9-C-SEG**]() | 640 | 27.4M | 145.5G | **53.3%** | **43.5%** |\n\n#### Panoptic Segmentation\n\n[`gelan-c-pan.pt`](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-pan.pt)\n\n`object detection` `instance segmentation` `semantic segmentation` `stuff segmentation` `panoptic segmentation`\n\n``` shell\n# coco\u002Flabels\u002F{split}\u002F*.txt\n# polygon (1 instance 1 line)\n# coco\u002Fstuff\u002F{split}\u002F*.txt\n# polygon (1 semantic 1 line)\npython panoptic\u002Ftrain.py --workers 8 --device 0 --batch 32  --data coco.yaml --img 640 --cfg models\u002Fpanoptic\u002Fgelan-c-pan.yaml --weights '' --name gelan-c-pan --hyp hyp.scratch-high.yaml --no-overlap --epochs 300 --close-mosaic 10\n```\n\n| Model | Test Size | Param. | FLOPs | AP\u003Csup>box\u003C\u002Fsup> | AP\u003Csup>mask\u003C\u002Fsup>  | mIoU\u003Csub>164k\u002F10k\u003C\u002Fsub>\u003Csup>semantic\u003C\u002Fsup> | mIoU\u003Csup>stuff\u003C\u002Fsup> | PQ\u003Csup>panoptic\u003C\u002Fsup> |\n| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |\n| [**GELAN-C-PAN**](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov9\u002Freleases\u002Fdownload\u002Fv0.1\u002Fgelan-c-pan.pt) | 640 | 27.6M | 146.7G | **52.6%** | **42.5%** | **39.0%\u002F48.3%** | **52.7%** | **39.4%** |\n| [**YOLOv9-C-PAN**]() | 640 | 28.8M | 187.0G | **52.7%** | **43.0%** | **39.8%\u002F-** | **52.2%** | **40.5%** |\n\n#### Image Captioning (not yet released)\n\n\u003C!--[`gelan-c-cap.pt`]()-->\n\n`object detection` `instance segmentation` `semantic segmentation` `stuff segmentation` `panoptic segmentation` `image captioning`\n\n``` shell\n# coco\u002Flabels\u002F{split}\u002F*.txt\n# polygon (1 instance 1 line)\n# coco\u002Fstuff\u002F{split}\u002F*.txt\n# polygon (1 semantic 1 line)\n# coco\u002Fannotations\u002F*.json\n# json (1 split 1 file)\npython caption\u002Ftrain.py --workers 8 --device 0 --batch 32  --data coco.yaml --img 640 --cfg models\u002Fcaption\u002Fgelan-c-cap.yaml --weights '' --name gelan-c-cap --hyp hyp.scratch-high.yaml --no-overlap --epochs 300 --close-mosaic 10\n```\n\n| Model | Test Size | Param. | FLOPs |  AP\u003Csup>box\u003C\u002Fsup> | AP\u003Csup>mask\u003C\u002Fsup>  | mIoU\u003Csub>164k\u002F10k\u003C\u002Fsub>\u003Csup>semantic\u003C\u002Fsup>  | mIoU\u003Csup>stuff\u003C\u002Fsup> | PQ\u003Csup>panoptic\u003C\u002Fsup> | BLEU@4\u003Csup>caption\u003C\u002Fsup> | CIDEr\u003Csup>caption\u003C\u002Fsup> |\n| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |\n| [**GELAN-C-CAP**]() | 640 | 47.5M | - | **51.9%** | **42.6%** | **42.5%\u002F-** | **56.5%** | **41.7%** | **38.8** | **122.3** |\n| [**YOLOv9-C-CAP**]() | 640 | 47.5M | - | **52.1%** | **42.6%** | **43.0%\u002F-** | **56.4%** | **42.1%** | **39.1** | **122.0** |\n\u003C!--| [**YOLOR-MT**]() | 640 | 79.3M | - | **51.0%** | **41.7%** | **-\u002F49.6%** | **55.9%** | **40.5%** | **35.7** | **112.7** |-->\n\n\n## Acknowledgements\n\n\u003Cdetails>\u003Csummary> \u003Cb>Expand\u003C\u002Fb> \u003C\u002Fsummary>\n\n* [https:\u002F\u002Fgithub.com\u002FAlexeyAB\u002Fdarknet](https:\u002F\u002Fgithub.com\u002FAlexeyAB\u002Fdarknet)\n* [https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolor](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolor)\n* [https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov7](https:\u002F\u002Fgithub.com\u002FWongKinYiu\u002Fyolov7)\n* [https:\u002F\u002Fgithub.com\u002FVDIGPKU\u002FDynamicDet](https:\u002F\u002Fgithub.com\u002FVDIGPKU\u002FDynamicDet)\n* [https:\u002F\u002Fgithub.com\u002FDingXiaoH\u002FRepVGG](https:\u002F\u002Fgithub.com\u002FDingXiaoH\u002FRepVGG)\n* [https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fyolov5](https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fyolov5)\n* [https:\u002F\u002Fgithub.com\u002Fmeituan\u002FYOLOv6](https:\u002F\u002Fgithub.com\u002Fmeituan\u002FYOLOv6)\n\n\u003C\u002Fdetails>\n","YOLOv9 是一个基于论文《YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information》实现的目标检测框架。该项目通过可编程梯度信息来优化学习过程，提供了多种不同规模的模型（如YOLOv9-T、YOLOv9-S等），在MS COCO数据集上表现出色，具有高精度和较低的参数量及计算量。项目支持自定义训练、ONNX导出以及TensorRT推理等功能，便于用户根据需求进行扩展和部署。适用于需要高效且准确目标检测的各种场景，例如自动驾驶、安防监控和工业检测等领域。",2,"2026-06-11 03:39:41","high_star"]