3D reconstruction open source code summary [keep updated]

3D reconstruction open source code summary , Irregular update .

Synchronize to Github Warehouse ：https://github.com/ethan-li-coding/Source-Codes-3DVison

SFM

• Openmvg【3.5k stars】【Mozilla Public License Version 2.0】【https://github.com/openMVG/openMVG】
  
  A multi view geometry 3D reconstruction algorithm library , Recover the three-dimensional pose of the camera from the disordered two-dimensional image set .SFM（Structure From Motion） Classic algorithm library .
  
  

• Colmap【3k stars】【new BSD license】【https://github.com/colmap/colmap】
  
  A multi view geometry 3D reconstruction algorithm library , Recover the three-dimensional pose of the camera from the disordered two-dimensional image set （SFM） And multi view stereo reconstruction （MVS）.
  
  

• AliceVision【1.8k stars】【MPL2 license】【https://github.com/alicevision/AliceVision】
  
  An open source photogrammetry system framework .
  
  

• TheiaSfM【600+k stars】【New BSD license】【https://github.com/sweeneychris/TheiaSfM】
  
  Google launched open source SFM Algorithm library .
  

• MVE【700+k stars】【BSD 3-Clause License】【https://github.com/simonfuhrmann/mve】
  
  

• OpenSfM【2.2k stars】【BSD license】【https://github.com/mapillary/OpenSfM】
  
  One use Python Realized SFM Algorithm library .
  

MVS

• openmvs【1.6k stars】【AGPLv3】【https://github.com/cdcseacave/openMVS】
  
  An open source algorithm library for multi view stereo reconstruction , Based on image set with pose information , Reconstruct a high-quality texture model , The effect is excellent .
  
  
• meshroom【6.5k stars】【MPL2 license】【https://github.com/alicevision/meshroom】
  
  and OpenMVS Like the library , It is also based on image sets with pose information , Reconstruct a high-quality texture model , But it didn't work OpenMVS good , Faster . With graphic interface , Well done .
  
  

SLAM

VSLAM

• orb-slam2【6.2k stars】【GPLv3 license】【https://github.com/raulmur/ORB_SLAM2】
  
  Very classic Monocular / Binocular /RGBD-slam Algorithm library
  
• orb-slam3【2.5k stars】【GPLv3 license】【https://github.com/UZ-SLAMLab/ORB_SLAM3】
  
  The first one can use monocular 、 Stereoscopic and RGB-D The camera , Use pinhole and fisheye lens models for vision 、 Visual inertia and multiple maps SLAM The system of .
  

  Monocular slam

  • orb-slam【1.1k stars】【GPLv3 license】【https://github.com/raulmur/ORB_SLAM】
    
    Very classic monocular slam Algorithm library
    

  RGBD slam

  • KinectFusion【280+ stars】【MIT License】【https://github.com/chrdiller/KinectFusionLib】
    
    be based on KinectFusion This paper implements an open source algorithm ,paper：KinectFusion: Real-time dense surface mapping and tracking
    

  • InfiniTAM∞ v3【680+ stars】【Oxford University Innovation Academic License】【https://github.com/victorprad/InfiniTAM】
    
    The open source multi platform real-time large-scale deep integration and tracking done by the Oxford University team , fast , stay Windows、Linux、Ios、Android Can be compiled and run , And the speed is ok .
    
    

  • VoxelHashing【460+ stars】【Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License】【https://github.com/niessner/VoxelHashing】
    
    be based on TSDF Classic depth fusion algorithm of voxel fusion , Yes CUDA edition , fast .
    

  • ElasticFusion【1.3k stars】【non-commercial use only】【https://github.com/mp3guy/ElasticFusion】
    
    One for the RGBD The data is based on Surfel Of Fusion Algorithm . Suitable for indoor reconstruction .
    
    

  • ElasticReconstruction【480+ stars】【MIT license】【https://github.com/qianyizh/ElasticReconstruction】
    
    One for the RGBD The data is based on Voxel Of Fusion Algorithm . Suitable for indoor reconstruction .
    
    

  • BundleFusion【1k stars】【non-commercial applications】【https://github.com/niessner/BundleFusion】
    
    A real-time and globally consistent 3D reconstruction algorithm library , be based on TSDF, The effect is now Fusion The best of all .
    
    

  • badslam【400+ stars】【BSD-3-Clause License】【CVPR2019】【https://github.com/ETH3D/badslam】
    
    One is based on RGBD Real time direct method of data BA-SLAM Algorithm , Suitable for outdoor reconstruction , From the Federal Institute of technology in Zurich （ETH） The team .
    
    

  • tsdf-fusion 【400+ stars】 【BSD-2-Clause License】【https://github.com/andyzeng/tsdf-fusion】
    
    One fuses multiple registered depth maps into TSDF Voxel Algorithm ,TSDF Voxels can be used for Marching Cubes Grid construction
    
    

Feature matching

• CudaSift【570+ stars】【non-commercial】【https://github.com/Celebrandil/CudaSift】
  
  One CUDA Fast implementation SIFT Algorithm ,SIFT It is a classic scale invariant feature matching algorithm
  
  

Stereo matching

• SemiglobalMatching（SGM）【400+ stars】【MIT License】【https://github.com/ethan-li-coding/SemiGlobalMatching】
  
  Classic semi global stereo matching algorithm SGM Algorithm library . High efficiency and strong adaptability .
  
  
• PatchMatchStereo【250+ stars】【MIT License】【https://github.com/ethan-li-coding/PatchMatchStereo】
  
  Classic inclined plane stereo matching algorithm library PatchMatch. It works great ！
  
  
  -AD-Cense 【160+ stars】【MIT License】【https://github.com/ethan-li-coding/AD-Census】
  
  classic AD-Census Stereo matching algorithm , Efficient , The effect is good .Intel RealSense D400 Stereo Modular algorithm .
  
  
  -MC-CNN【600+ stars】【BSD-2-Clause License】【https://github.com/jzbontar/mc-cnn】
  
  The classic work of deep learning stereo matching algorithm .
  
  

Point cloud splicing

• ICPCUDA【400+ stars】【No License】【https://github.com/mp3guy/ICPCUDA】
  
  A depth map based data （DepthMap） Fast ICP Splice Library （CUDA Speed up ）, about VGA Size （640x480）, stay NVIDIA GeForce GTX TITAN X Up to 750Hz The speed of . The author also opened source KinectFusion.
  
• CUDA-ScanMatcher-ICP [100+ stars] [MIT License]【 https://github.com/botforge/CUDA-ScanMatcher-ICP】
  
  One is based on CUDA Of ICP Point cloud splicing algorithm , It is suitable for laser scanning point cloud and general point cloud .
  
  
• Super4PCS【300+ stars】【Apache License V2】【https://github.com/nmellado/Super4PCS】
  
  Point cloud rough matching algorithm Super4PCS The implementation of the .
  
  

Grid construction

• PoissonRecon 【700+ stars】【MIT License】【https://github.com/mkazhdan/PoissonRecon】
  
  Poisson reconstruction , A grid construction algorithm , The input is an unordered point cloud with normals .
  
  

Point cloud mesh processing

• pcl【6.3k stars】【BSD License】【https://github.com/PointCloudLibrary/pcl】
  
  This is probably the most widely used point cloud processing algorithm library , You can use it to display pictures 、 Show point cloud , It also implements many point cloud processing algorithms , Such as point cloud filtering 、 Point cloud registration 、 Point cloud segmentation , Its official website is https://pointclouds.org/.
  
  
• OpenMesh【BSD 3 clause license】【https://www.graphics.rwth-aachen.de/software/openmesh/】
  
  A very classic grid processing library , Based on half data structure , It's very efficient , And very stable , There are many implementations of basic grid algorithms .
  
  
• vcglib【560+ stars】【GPLv3.0 License】【http://vcg.isti.cnr.it/vcglib/】
  
  and OpenMesh The same grid processing basic algorithm library , Also very efficient and stable , Well known grid processing software MeshLib Just use it .
  
  

Texture mapping

• mvs-texturing【500+ stars】【BSD 3-Clause License】【https://github.com/nmoehrle/mvs-texturing】
  
  A texture mapping algorithm library for multi view stereo reconstruction ,OpenMVS、MVE This algorithm is used to do texture mapping .
  
  
• UVAtlas【500+ stars】【MIT License】【https://github.com/microsoft/UVAtlas】
  
  Microsoft open source automatic UV Expansion algorithm ,UV Expansion can be used to map meshes to a two-dimensional UV On the drawing , A mesh mapped texture can be used UV Figure to organize , Improve the utilization of texture map .
  
  

3D rendering

• potree 【2.3k stars】【FreeBSD Licnese】【https://github.com/potree/potree】
  
  One is based on WebGL Large point cloud rendering project , It is interesting to .
  
  
• OpenSceneGraph（OSG）【2.2k stars】【 OpenSceneGraph Public License】【https://github.com/openscenegraph/OpenSceneGraph】
  
  A very mature and easy-to-use cross platform 3D data rendering Library , Rendering includes point clouds 、 grid 、 3D data including texture model , Suitable for large-scale software as a rendering engine , It can also be handled LOD data .
  
  
• Pangolin【1.3k stars】【MIT License】【https://github.com/stevenlovegrove/Pangolin】
  
  A lightweight cross platform 3D rendering Library , Can be used for slam Rendering engine for related projects .