{"id":19187445,"url":"https://github.com/materight/non-rigid-object-tracking","last_synced_at":"2026-06-15T20:31:22.500Z","repository":{"id":93569071,"uuid":"334694647","full_name":"materight/non-rigid-object-tracking","owner":"materight","description":"A set of algorithms for non-rigid tracking of multiple objects in videos from different domains.","archived":false,"fork":false,"pushed_at":"2021-04-09T15:25:44.000Z","size":18517,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-09-12T19:01:01.409Z","etag":null,"topics":["computer-vision","object-tracking","opencv"],"latest_commit_sha":null,"homepage":"","language":"C++","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/materight.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,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2021-01-31T15:58:00.000Z","updated_at":"2021-09-13T18:50:00.000Z","dependencies_parsed_at":"2023-03-21T00:46:58.902Z","dependency_job_id":null,"html_url":"https://github.com/materight/non-rigid-object-tracking","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/materight/non-rigid-object-tracking","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/materight%2Fnon-rigid-object-tracking","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/materight%2Fnon-rigid-object-tracking/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/materight%2Fnon-rigid-object-tracking/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/materight%2Fnon-rigid-object-tracking/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/materight","download_url":"https://codeload.github.com/materight/non-rigid-object-tracking/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/materight%2Fnon-rigid-object-tracking/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34379915,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-15T02:00:07.085Z","response_time":63,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"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":["computer-vision","object-tracking","opencv"],"created_at":"2024-11-09T11:19:05.034Z","updated_at":"2026-06-15T20:31:22.486Z","avatar_url":"https://github.com/materight.png","language":"C++","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Non-rigid Multi-object Tracking\n**Azzolin Steve, Destro Matteo** \\\n**Signal, Image and Video** \\\n**Year 2020/21**\n\n\n## Get started\n\n### Import the conda env\n- Run `conda env create -f environment.yaml`.\n- Run `conda activate non_rigid_tracking`.\n- **Note:** In case you want to run the Lin-pun tracker, you also need to compile some additional C libraries by running `cd prim \u0026\u0026 make`.\n\n### Run sample\nSimply run `main.py` to run the algorithm with the default parameters and configurations.\n\n### Configurations\n`config.yaml` contains all the configurable parameters for the maskers. The most important ones are:\n- `input_video`: path of the video to use, relative to the root directory of the project.\n- `masker`: identifier of the masker to be used, see section [Maskers](#maskers) for a list of the available maskers.\n- `manual_roi_selection`: set to False to use the polygon specified in the `pts` parameter as initial selection. If set to True, the algorithm asks for a manual selection. \n- `show_masks`: Set to True to show the resulting mask while running the algorithm.\n\nSee the comments in `config.yaml` for more details and a description of the other parameters.\n\n\n\n\n## Project structure\n\n### Maskers\nThe folder `masker` contains separate classes for each algorithm proposal. The available maskers are:\n- **BgSub** (`bg_subtractor_masker.py`): background subtractor.\n- **LinPuntracker** (`lin_pun_tracker.py`): Lin-pun highly non-rigid object tracker. Please note that this algorithm was not thoroughly tested.\n- **OpticalFlow** (`optical_flow_masker.py`): Optical Flow and Convex Hull masker (OFCH).\n- **PC** (`pixel_classification.py`): Pixel Classification (PC) masker.\n- **GrabCut** (`grab_cut.py`): GrabCut-based algorithm.\n\n### Input/Output\nThe videos used for testing the algorithms are available in the `Input` folder. These videos were taken from the [Seg Track V2](https://web.engr.oregonstate.edu/~lif/SegTrack2/dataset.html) public dataset.\n\nAfter running the algorithm, the resulting mask can be found inside `Output` (or in the custom folder specified in `config.yaml`).\n\n### Benchmark\n`benchmark.py` is a utility script to easily test different parameters combination automatically. It is useful in particular with the *PixelClassification* tracker which has many hyper-parameters.\n\nThe benchmark uses `config_benchmark.yaml` as config file (see section [Configurations](#configurations) for more details). For hyper-parameter testing, inside of `benchmark.py` two variables can be found:\n- `VIDEOS`: the file names of the video files we want to run the benchmark on. Note that if you want to add a new video to the benchmark, you need to define the points of the selection masks in `polygons.yaml`.\n- `HYPERPARAMS`: contains, for each hyper-parameter, a list of values to be tested, for which the script automatically generates and tests any possible combination.\n\n\nThe results are then saved into `benchmark_results.csv`, with the obtained benchmark and the processing time required for each video.\n\n## Results\nThe following table show the results for the three best approaches proposed. The benchmark score was computed with the intersection over union (IoU) metric.\n\n| Sequence  | OFCH | PC   | GrabCut |\n| --------- | ---- | ---- | ------- |\n| Parachute | 0.60 | 0.77 |  0.79   |\n| Soldier   | 0.58 | 0.74 |  0.75   |\n| Worm      | 0.35 | 0.72 |  0.73   |\n| Frog      | 0.32 | 0.76 |  0.61   |\n\n\n\\\nBelow an example of the resulting mask for each algorithm:\n\n*Optical Flow and Convex Hull (OFCH)*  \\\n![Optical Flow and Convex Hull mask example](img/OFCH_soldier.gif)\n\n \n*Pixel Classification (PC)* \\\n![Pixel Classification mask example](img/PC_soldier.gif)\n\n*GrabCut* \\\n![GrabCut mask example](img/GrabCut_soldier.gif)\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmateright%2Fnon-rigid-object-tracking","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmateright%2Fnon-rigid-object-tracking","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmateright%2Fnon-rigid-object-tracking/lists"}