{"id":18008950,"url":"https://github.com/mp3guy/icpcuda","last_synced_at":"2025-04-04T18:07:51.044Z","repository":{"id":38869907,"uuid":"32736402","full_name":"mp3guy/ICPCUDA","owner":"mp3guy","description":"Super fast implementation of ICP in CUDA for compute capable devices 3.5 or higher","archived":false,"fork":false,"pushed_at":"2021-10-06T11:14:47.000Z","size":91,"stargazers_count":713,"open_issues_count":17,"forks_count":180,"subscribers_count":32,"default_branch":"master","last_synced_at":"2025-03-28T17:08:24.036Z","etag":null,"topics":["cuda","icp"],"latest_commit_sha":null,"homepage":"","language":"C++","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"bsd-3-clause","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/mp3guy.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}},"created_at":"2015-03-23T14:17:21.000Z","updated_at":"2025-03-27T02:27:56.000Z","dependencies_parsed_at":"2022-08-01T00:08:20.785Z","dependency_job_id":null,"html_url":"https://github.com/mp3guy/ICPCUDA","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/mp3guy%2FICPCUDA","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mp3guy%2FICPCUDA/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mp3guy%2FICPCUDA/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mp3guy%2FICPCUDA/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/mp3guy","download_url":"https://codeload.github.com/mp3guy/ICPCUDA/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247226215,"owners_count":20904465,"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":["cuda","icp"],"created_at":"2024-10-30T02:08:18.504Z","updated_at":"2025-04-04T18:07:51.008Z","avatar_url":"https://github.com/mp3guy.png","language":"C++","funding_links":[],"categories":[],"sub_categories":[],"readme":"# ICPCUDA\nSuper fast implementation of ICP in CUDA for compute capable devices 3.5 or higher. On an NVIDIA GeForce GTX TITAN X it runs at over __750Hz__ (using projective data assocation). Last tested with Ubuntu 18.04.2, CUDA 10.1 and NVIDIA drivers 418.39.\n\nRequires CUDA, includes [Pangolin](https://github.com/stevenlovegrove/Pangolin), [Eigen](https://github.com/stevenlovegrove/eigen) and [Sophus](https://github.com/stevenlovegrove/Sophus) third party submodules. I've built it to take in raw TUM RGB-D datasets to do frame-to-frame dense ICP as an example application.\n\nInstall;\n\n```bash\nsudo apt-get install build-essential cmake libglew-dev libpng-dev\ngit clone https://github.com/mp3guy/ICPCUDA.git\ncd ICPCUDA\ngit submodule update --init\ncd third-party/Pangolin/\nmkdir build\ncd build/\ncmake ../ -DEIGEN_INCLUDE_DIR=\u003cabsolute_path_to_Eigen_submodule\u003e\nmake -j12\ncd ../../../\nmkdir build\ncd build/\ncmake ..\nmake -j12\n```\n\nThe particular version of ICP implemented is the one introduced by [KinectFusion](http://homes.cs.washington.edu/~newcombe/papers/newcombe_etal_ismar2011.pdf). This means a three level coarse-to-fine registration pyramid, from 160x120 to 320x240 and finally 640x480 image sizes, with 4, 5 and 10 iterations per level respectively. \n\nRun like;\n\n```bash\n./ICP ~/Desktop/rgbd_dataset_freiburg1_desk/ -v\n```\n\nWhere ~/Desktop/rgbd\\_dataset\\_freiburg1\\_desk/ contains the depth.txt file, for more information see [here](http://vision.in.tum.de/data/datasets/rgbd-dataset).\n\nThe main idea to getting the best performance is determining the best thread/block sizes to use. I have provided an exhaustive search function to do this, since it varies between GPUs. Simply pass the \"-v\" switch to the program to activate the search. The code will then first do a search for the best thread/block sizes and then run ICP and output something like this on an nVidia GeForce GTX TITAN X;\n\n```bash\nGeForce GTX TITAN X\nSearching for the best thread/block configuration for your GPU...\nBest: 256 threads, 96 blocks (1.3306ms), 100%\nICP: 1.3236ms\nICP speed: 755Hz\n```\n\nThe code will output one file; output.poses. You can evaluate it on the TUM benchmark by using their tools. I get something like this;\n\n```bash\npython ~/stuff/Kinect_Logs/Freiburg/evaluate_ate.py ~/Desktop/rgbd_dataset_freiburg1_desk/groundtruth.txt output.poses \n0.144041\n```\n\nThe difference in values comes down to the fact that each method uses a different reduction scheme and floating point operations are [not associative](https://halshs.archives-ouvertes.fr/hal-00949355v1/document).\n\nAlso, if you're using this code in academic work and it would be suitable to do so, please consider referencing some of my possibly relevant [research](http://www.thomaswhelan.ie/#publications) in your literature review/related work section. \n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmp3guy%2Ficpcuda","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmp3guy%2Ficpcuda","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmp3guy%2Ficpcuda/lists"}