{"id":13441658,"url":"https://github.com/arangesh/TrackMPNN","last_synced_at":"2025-03-20T12:32:14.334Z","repository":{"id":40550710,"uuid":"201290589","full_name":"arangesh/TrackMPNN","owner":"arangesh","description":null,"archived":false,"fork":false,"pushed_at":"2022-05-03T02:13:44.000Z","size":2043,"stargazers_count":8,"open_issues_count":8,"forks_count":1,"subscribers_count":4,"default_branch":"master","last_synced_at":"2024-07-31T03:30:11.201Z","etag":null,"topics":["graph-attention-networks","graph-neural-networks","message-passing","multi-object-tracking"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/arangesh.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2019-08-08T15:53:19.000Z","updated_at":"2024-06-04T06:01:50.000Z","dependencies_parsed_at":"2022-08-09T23:00:18.546Z","dependency_job_id":null,"html_url":"https://github.com/arangesh/TrackMPNN","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/arangesh%2FTrackMPNN","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arangesh%2FTrackMPNN/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arangesh%2FTrackMPNN/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arangesh%2FTrackMPNN/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/arangesh","download_url":"https://codeload.github.com/arangesh/TrackMPNN/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":221760095,"owners_count":16876351,"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":["graph-attention-networks","graph-neural-networks","message-passing","multi-object-tracking"],"created_at":"2024-07-31T03:01:36.608Z","updated_at":"2025-03-20T12:32:14.326Z","avatar_url":"https://github.com/arangesh.png","language":"Python","funding_links":[],"categories":["Python","算法论文"],"sub_categories":["**2021**"],"readme":"# TrackMPNN: A Message Passing Neural Network for End-to-End Multi-object Tracking\n\nThis is the Pytorch implementation of TrackMPNN for the KITTI and BDD100K multi-object tracking (MOT) datasets.\n\n## Installation\n1) Clone this repository\n2) Install Pipenv:\n```shell\npip3 install pipenv\n```\n3) Install all requirements and dependencies in a new virtual environment using Pipenv:\n```shell\ncd TrackMPNN\npipenv install\n```\n4) Get link for desired PyTorch wheel from [here](https://download.pytorch.org/whl/torch_stable.html) and install it in the Pipenv virtual environment as follows:\n```shell\npipenv install https://download.pytorch.org/whl/cu100/torch-1.2.0-cp36-cp36m-manylinux1_x86_64.whl\n```\n5) Clone and make DCNv2 (note gcc-8 is highest supported version incase you're on ubuntu 20.04 +):\n```shell\ncd models/dla\ngit clone https://github.com/CharlesShang/DCNv2\ncd DCNv2\n./make.sh\n```\n6) Download the [imagenet pre-trained embedding network weights](https://drive.google.com/file/d/1fwnWaXftLBBARN_CG-BmzrBNfMNQaiAV/view?usp=sharing) to the `TrackMPNN/weights` folder.\n\n## KITTI\n### Dataset\n1) Download and extract the KITTI multi-object tracking (MOT) dataset (including images, labels, and calibration files).\n2) Download the [RRC and CenterTrack detections](https://drive.google.com/file/d/1PIDr9GcTayXw7GtmQ_R7IMm4YiT1eU_W/view?usp=sharing) for both `training` and `testing` splits and add them to the KITTI MOT folder. The dataset should be organized as follows:\n```plain\n└── kitti-mot\n    ├── training/\n    |   └── calib/\n    |   └── image_02/\n    |   └── label_02/\n    |   └── rrc_detections/\n    |   └── centertrack_detections/\n    └── testing/\n        └── calib/\n        └── image_02/\n        └── rrc_detections/\n        └── centertrack_detections/\n```\n\n### Training\nTrackMPNN can be trained using RRC detections as follows:\n```shell\npipenv shell # activate virtual environment\npython train.py --dataset=kitti --dataset-root-path=/path/to/kitti-mot/ --cur-win-size=5 --detections=rrc --feats=2d --category=Car --no-tp-classifier --epochs=30 --random-transforms\nexit # exit virtual environment\n```\nTrackMPNN can also be trained using CenterTrack detections as follows:\n```shell\npipenv shell # activate virtual environment\npython train.py --dataset=kitti --dataset-root-path=/path/to/kitti-mot/ --cur-win-size=5 --detections=centertrack --feats=2d --category=All --no-tp-classifier --epochs=50 --random-transforms\nexit # exit virtual environment\n```\nBy default, the model is trained to track `All` object categories, but you can supply the `--category` argument with any one of the following options: `['Pedestrian', 'Car', 'Cyclist', 'All']`.\n\n### Inference\nInference on the `testing` split can be carried out using [this](https://github.com/arangesh/TrackMPNN/blob/master/infer.py) script as follows:\n```shell\npipenv shell # activate virtual environment\npython infer.py --snapshot=/path/to/snapshot.pth --dataset-root-path=/path/to/kitti-mot/ --hungarian\nexit # exit virtual environment\n```\nAll settings from training will be carried over for inference.\n\nConfig files, logs, results and snapshots from running the above scripts will be stored in the `TrackMPNN/experiments` folder by default.\n\n\n### Visualizing Inference Results \nYou can use the `utils/visualize_mot.py` script to generate a video of the tracking results after running the inference script:\n```shell\npipenv shell # activate virtual environment\npython utils/visualize_mot.py /path/to/testing/inference/results /path/to/kitti-mot/testing/image_02\nexit\n```\nThe videos will be stored in the same folder as the inference results.\n\n## BDD100K\n### Dataset\n1) Download and extract the BDD100K multi-object tracking (MOT) dataset (including images, labels, and calibration files).\n2) Download the [HIN and Libra detections](https://drive.google.com/file/d/1PIDr9GcTayXw7GtmQ_R7IMm4YiT1eU_W/view?usp=sharing) for `training`, `validation` and `testing` splits and add them to the BDD100K MOT folder. The dataset should be organized as follows:\n```plain\n└── bdd100k-mot\n    ├── training/\n    |   └── calib/\n    |   └── image_02/\n    |   └── label_02/\n    |   └── hin_detections/\n    |   └── libra_detections/\n    ├── validation/\n    |   └── calib/\n    |   └── image_02/\n    |   └── label_02/\n    |   └── hin_detections/\n    |   └── libra_detections/\n    └── testing/\n        └── calib/\n        └── image_02/\n        └── hin_detections/\n        └── libra_detections/\n```\n\n### Training\nTrackMPNN can be trained using HIN detections as follows:\n```shell\npipenv shell # activate virtual environment\npython train.py --dataset=bdd100k --dataset-root-path=/path/to/bdd100k-mot/ --cur-win-size=5 --detections=libra --feats=2d --category=All --no-tp-classifier --epochs=20  --random-transforms\nexit # exit virtual environment\n```\nBy default, the model is trained to track `All` object categories, but you can supply the `--category` argument with any one of the following options: `['pedestrian', 'rider', 'car', 'bus', 'truck', 'train', 'motorcycle', 'bicycle', 'All']`.\n\n### Inference\nInference on the `testing` split can be carried out using [this](https://github.com/arangesh/TrackMPNN/blob/master/infer.py) script as follows:\n```shell\npipenv shell # activate virtual environment\npython infer.py --snapshot=/path/to/snapshot.pth --dataset-root-path=/path/to/bdd100k-mot/ --hungarian\nexit # exit virtual environment\n```\nAll settings from training will be carried over for inference.\n\nConfig files, logs, results and snapshots from running the above scripts will be stored in the `TrackMPNN/experiments` folder by default.\n\n\n### Visualizing Inference Results \nYou can use the `utils/visualize_mot.py` script to generate a video of the tracking results after running the inference script:\n```shell\npipenv shell # activate virtual environment\npython utils/visualize_mot.py /path/to/testing/inference/results /path/to/bdd100k-mot/testing/image_02\nexit\n```\nThe videos will be stored in the same folder as the inference results.","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Farangesh%2FTrackMPNN","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Farangesh%2FTrackMPNN","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Farangesh%2FTrackMPNN/lists"}