[{"data":1,"prerenderedAt":493},["ShallowReactive",2],{"method-voxelhashing2013":3},{"method":4,"reference":46,"equipment":69,"figures":90,"results":91},{"id":5,"label":6,"shortName":7,"title":8,"year":9,"era":10,"cluster":11,"scope":12,"keyIdeaZh":13,"keyIdeaEn":14,"fulltextStatus":15,"publicationStatus":16,"recommendation":17,"constructionRelevance":18,"validationEnvironment":19,"strengths":21,"limitations":24,"sensors":26,"platform":28,"estimator":30,"association":31,"timeModel":32,"deskew":33,"loopClosure":34,"globalOptimization":35,"mapRepresentation":36,"prior":37,"outputGeometry":38,"compute":39,"codeUrl":40,"codeLicense":41,"relatedVersions":42},"voxelhashing2013","Nießner et al., 2013","Voxel Hashing","Real-time 3D reconstruction at scale using voxel hashing",2013,"classic","C08","map_representation_or_reconstruction","體素雜湊（voxel hashing）以簡單的空間雜湊表只在有量測的表面附近配置 TSDF 體素區塊，避免規則網格或階層式資料結構的記憶體負擔。資料可在 GPU 與主機之間串流進出雜湊表，讓感測器移動時重建範圍可擴大。它主要是地圖表示與融合的資料結構，而非完整 SLAM。","Voxel hashing stores TSDF voxel blocks in a spatial hash only where surfaces are observed and streams them between GPU and host, enabling large-scale real-time volumetric fusion.","full_text_reviewed","peer_reviewed_published","background","論文未報告營建測試。其稀疏體素儲存對大範圍室內或構造物稠密重建的記憶體可行性有關，但幾何精度仍取決於位姿與深度感測器誤差（推論）。",[20],"completed_building",[22,23],"Reconstructs both fine details and large-scale environments in real time (abstract)","Data can be streamed in and out of the hash table for scalability (abstract)",[25],"keep; add: active infrared depth quality degrades outdoors (Sec. 9); voxels below 2 mm gave no visible improvement because of depth sensor limits (Sec. 9); ray-based rather than frustum-based block allocation is an approximation chosen for speed (Sec. 5)",[27],"RGB-D",[29],"not stated explicitly; live captures with a Kinect for Windows camera or an Asus Xtion (both RGB-D at 30 Hz) moved by a user (Sec. 5 mentions 'the mobility of the user'; Sec. 9)","frame-to-model point-to-plane ICP against the raycast surface with projective data association, linearised on the GPU and solved by SVD on the CPU (camera tracking paragraph)","projective data association for point-plane ICP, with an optional colour weighting term (camera tracking paragraph)","discrete poses","not_reported","none (authors state that no drift correction is explicitly handled; results section)","none reported in sections read","TSDF in 8x8x8 voxel blocks (8 bytes per voxel: SDF, RGB, weight) indexed by a spatial hash table of 2^21 entries with bucket size 2; blocks outside an active sphere of 8 m radius centred 4 m in front of the camera streamed to host memory in 1 m^3 chunks and streamed back when revisited (Sec. 4, 8, 9.1)","none","TSDF with per-voxel colour and weight; isosurface extracted by raycasting for tracking and display, and the authors state isosurfaces can be extracted by raycasting or polygonisation; output meshes are shown in Fig. 10 (Sec. 3, 7, Fig. 10)","DirectX 11 compute shaders on Intel Core i7 3.4 GHz, 16 GB RAM and one NVIDIA GeForce GTX Titan; 21.8 ms average per frame (about 46 fps) including 8.0 ms ICP; 34 MB hash table, 1 GB GPU heap, under 300 MB surface data versus well over 5 GB for a regular grid at 8 mm (Sec. 9, 9.1)","https:\u002F\u002Fgithub.com\u002Fniessner\u002FVoxelHashing","CC BY-NC-SA 3.0 (LICENSE.txt)",[43],{"relation":44,"title":45,"doi_or_url":40},"code_release","VoxelHashing (Depth Sensing and Voxel Hashing)",{"id":5,"kind":47,"shortName":7,"title":8,"authors":48,"year":9,"venue":53,"venueType":54,"publisher":55,"volumeIssuePages":56,"doi":57,"arxivId":58,"url":59,"firstPublicDate":60,"publicationStatus":16,"metadataStatus":61,"fulltextStatus":15,"era":10,"classicReason":62,"codeUrl":40,"cluster":11,"topics":63,"mdpi":65,"verification":66,"label":6,"fulltextRoute":67,"versionRead":68,"addedByCensus":65},"component",[49,50,51,52],"Matthias Nießner","Michael Zollhöfer","Shahram Izadi","Marc Stamminger","ACM Transactions on Graphics","journal","ACM","32(6):1-11 (Crossref pages)","10.1145\u002F2508363.2508374",null,"https:\u002F\u002Fniessnerlab.org\u002Fpapers\u002F2013\u002F4hashing\u002Fniessner2013hashing.pdf","2013-11","metadata_verified","principle reused: sparse spatial hashing of TSDF voxel blocks with GPU-CPU streaming removes the fixed-volume limit of KinectFusion and is used by BundleFusion.",[11,64],"C12",false,"corrected","author copy","author-hosted copy (niessnerlab.org, 11 pages, ACM SIGGRAPH template); version of record ACM TOG 32(6), Article 169, pp. 1-11, Copyright 2013 ACM; VoR PDF not compared (curl HTTP 403; no HTML full text on ACM DL)",[70,76,80,86],{"category":71,"model":72,"canonical":72,"role":73,"dataset":58,"specs":74,"locator":75},"rgbd","Kinect for Windows","method input","RGB-D data at 30 Hz; depth range assumed up to 8 m for streaming","Sec. 8, Sec. 9",{"category":71,"model":77,"canonical":77,"role":73,"dataset":58,"specs":78,"locator":79},"Asus Xtion","RGB-D data at 30 Hz; used for the scenes in Fig. 10","Sec. 9",{"category":81,"model":82,"canonical":82,"role":83,"dataset":58,"specs":84,"locator":85},"compute","Intel Core i7 3.4GHz CPU","compute for runtime","16 GB RAM","Sec. 9.1",{"category":81,"model":87,"canonical":87,"role":83,"dataset":58,"specs":88,"locator":89},"NVIDIA GeForce GTX Titan","single GPU; DirectX 11 compute shaders","Sec. 9, Sec. 9.1",[],{"totalRows":92,"groupCount":93,"groups":94,"others":484},24,5,[95,181,300,409],{"slug":96,"group":97,"sourceId":5,"sourceLabel":6,"table":98,"selfRows":99,"metrics":100,"seqs":123,"entrants":137,"cells":141,"outcomes":171,"locators":173,"hardware":174,"wordings":176,"notes":177},"voxelhashing2013-text-sec-9-1","voxelhashing2013:Text Sec. 9.1","Text Sec. 9.1",11,[101,105,107,109,111,113,116,119,121],{"label":102,"unit":103,"statistic":104,"alignment":37},"average time: entire pipeline (about 46 fps)","ms","mean",{"label":106,"unit":103,"statistic":104,"alignment":37},"average time: ICP pose estimation (37 %)",{"label":108,"unit":103,"statistic":104,"alignment":37},"average time: surface integration (21 %)",{"label":110,"unit":103,"statistic":104,"alignment":37},"average time: surface extraction and shading (22 %)",{"label":112,"unit":103,"statistic":104,"alignment":37},"average time: streaming and input data processing (20 %)",{"label":114,"unit":115,"statistic":33,"alignment":37},"hash table and auxiliary buffers","MB",{"label":117,"unit":118,"statistic":33,"alignment":37},"pre-allocated GPU heap for voxel blocks","GB",{"label":120,"unit":115,"statistic":33,"alignment":37},"surface data allocated on average at 8 mm voxels (reported as 'less than 300 MB'; less than 600 MB with colour)",{"label":122,"unit":103,"statistic":33,"alignment":37},"frame time",[124,128,130,133,135],{"dataset":125,"sequence":126,"environment":127},"own live captures","all test scenes (average)","indoor and outdoor real scenes",{"dataset":125,"sequence":129,"environment":127},"all test scenes",{"dataset":125,"sequence":131,"environment":132},"STATUES, hash table 2^21 entries (standard)","museum corridor with statues",{"dataset":125,"sequence":134,"environment":132},"STATUES, hash table 200K entries",{"dataset":125,"sequence":136,"environment":132},"STATUES, hash table 160K entries",[138],{"name":139,"methodId":5,"linkable":140,"proposed":140,"self":140},"Voxel hashing (proposed)",true,[142,146,149,152,155,158,160,162,165,167,169],[143,143,143,144,145,143,143,145,143],0,21.8,-1,[143,147,143,148,145,143,143,145,143],1,8,[143,150,143,151,145,143,143,145,143],2,4.6,[143,153,143,154,145,143,143,145,143],3,4.8,[143,156,143,157,145,143,143,145,143],4,4.4,[143,93,147,159,145,143,143,145,147],34,[143,161,147,147,145,143,143,145,147],6,[143,163,147,164,143,143,143,145,147],7,300,[143,148,150,166,145,143,143,145,150],21,[143,148,153,168,145,143,143,145,150],24.8,[143,148,156,170,145,143,143,145,150],25.6,[172],"upper bound: reported as less than 300 MB on average",[85],[175],"Intel Core i7 3.4 GHz CPU, 16 GB RAM, single NVIDIA GeForce GTX Titan (Sec. 9.1)",[],[178,179,180],"Average over all live test scenes (Kinect for Windows or Asus Xtion, 30 Hz) of the entire pipeline including display rendering; ICP with 15 iterations","Memory footprint of the data structure in the live test scenes","STATUES scene; effect of hash table size on occupancy and frame time (values given as approximate)",{"slug":182,"group":183,"sourceId":184,"sourceLabel":185,"table":186,"selfRows":156,"metrics":187,"seqs":192,"entrants":203,"cells":226,"outcomes":294,"locators":295,"hardware":296,"wordings":297,"notes":298},"bundlefusion2017-table-3","bundlefusion2017:Table 3","bundlefusion2017","Dai et al., 2017a","Table 3",[188],{"label":189,"unit":190,"statistic":191,"alignment":33},"ATE RMSE","cm","RMSE",[193,197,199,201],{"dataset":194,"sequence":195,"environment":196},"ICL-NUIM","kt0","synthetic living room",{"dataset":194,"sequence":198,"environment":196},"kt1",{"dataset":194,"sequence":200,"environment":196},"kt2",{"dataset":194,"sequence":202,"environment":196},"kt3",[204,206,208,210,213,215,218,220,222,224],{"name":205,"methodId":58,"linkable":65,"proposed":65,"self":65},"DVO SLAM",{"name":207,"methodId":58,"linkable":65,"proposed":65,"self":65},"RGB-D SLAM",{"name":209,"methodId":58,"linkable":65,"proposed":65,"self":65},"MRSMap",{"name":211,"methodId":212,"linkable":140,"proposed":65,"self":65},"Kintinuous","kintinuous2015",{"name":214,"methodId":5,"linkable":140,"proposed":65,"self":140},"VoxelHashing",{"name":216,"methodId":217,"linkable":140,"proposed":65,"self":65},"Elastic Fusion","elasticfusion2015",{"name":219,"methodId":58,"linkable":65,"proposed":65,"self":65},"Redwood (rigid)",{"name":221,"methodId":184,"linkable":140,"proposed":65,"self":65},"BundleFusion ablation: Ours (s), sparse only",{"name":223,"methodId":184,"linkable":140,"proposed":65,"self":65},"BundleFusion ablation: Ours (sd), sparse and local dense",{"name":225,"methodId":184,"linkable":140,"proposed":140,"self":65},"BundleFusion (Ours)",[227,229,231,233,235,237,239,241,243,245,247,249,251,253,255,256,258,260,262,263,265,267,268,269,271,272,273,275,277,278,280,282,283,284,285,287,288,291,292,293],[143,143,143,228,145,143,145,145,143],10.4,[143,143,147,230,145,143,145,145,143],2.9,[143,143,150,232,145,143,145,145,143],19.1,[143,143,153,234,145,143,145,145,143],15.2,[147,143,143,236,145,143,145,145,143],2.6,[147,143,147,238,145,143,145,145,143],0.8,[147,143,150,240,145,143,145,145,143],1.8,[147,143,153,242,145,143,145,145,143],43.3,[150,143,143,244,145,143,145,145,143],20.4,[150,143,147,246,145,143,145,145,143],22.8,[150,143,150,248,145,143,145,145,143],18.9,[150,143,153,250,145,143,145,145,143],109,[153,143,143,252,145,143,145,145,143],7.2,[153,143,147,254,145,143,145,145,143],0.5,[153,143,150,147,145,143,145,145,143],[153,143,153,257,145,143,145,145,143],35.5,[156,143,143,259,145,143,145,145,143],1.4,[156,143,147,261,145,143,145,145,143],0.4,[156,143,150,240,145,143,145,145,143],[156,143,153,264,145,143,145,145,143],12,[93,143,143,266,145,143,145,145,143],0.9,[93,143,147,266,145,143,145,145,143],[93,143,150,259,145,143,145,145,143],[93,143,153,270,145,143,145,145,143],10.6,[161,143,143,170,145,143,145,145,143],[161,143,147,153,145,143,145,145,143],[161,143,150,274,145,143,145,145,143],3.3,[161,143,153,276,145,143,145,145,143],6.1,[163,143,143,266,145,143,145,145,143],[163,143,147,279,145,143,145,145,143],1.2,[163,143,150,281,145,143,145,145,143],1.3,[163,143,153,281,145,143,145,145,143],[148,143,143,238,145,143,145,145,143],[148,143,147,254,145,143,145,145,143],[148,143,150,286,145,143,145,145,143],1.1,[148,143,153,279,145,143,145,145,143],[289,143,143,290,145,143,145,145,143],9,0.6,[289,143,147,261,145,143,145,145,143],[289,143,150,290,145,143,145,145,143],[289,143,153,286,145,143,145,145,143],[],[186],[],[],[299],"ICL-NUIM living-room trajectories kt0 to kt3 with synthetic noise; ATE RMSE; comparator values match those printed in ElasticFusion Table II; Ours (s) sparse-only and Ours (sd) sparse plus local dense are ablations; Redwood runs offline without colour",{"slug":301,"group":302,"sourceId":184,"sourceLabel":185,"table":303,"selfRows":156,"metrics":304,"seqs":306,"entrants":317,"cells":333,"outcomes":402,"locators":404,"hardware":405,"wordings":406,"notes":407},"bundlefusion2017-table-4","bundlefusion2017:Table 4","Table 4",[305],{"label":189,"unit":190,"statistic":191,"alignment":33},[307,311,313,315],{"dataset":308,"sequence":309,"environment":310},"TUM RGB-D","fr1\u002Fdesk","small scenes with simple camera trajectories; hand-held Kinect sequences with motion-capture ground truth (Sec. 6)",{"dataset":308,"sequence":312,"environment":310},"fr2\u002Fxyz",{"dataset":308,"sequence":314,"environment":310},"fr3\u002Foffice",{"dataset":308,"sequence":316,"environment":310},"fr3\u002Fnst",[318,319,320,321,322,323,324,327,329,330,331,332],{"name":205,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":207,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":209,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":211,"methodId":212,"linkable":140,"proposed":65,"self":65},{"name":214,"methodId":5,"linkable":140,"proposed":65,"self":140},{"name":216,"methodId":217,"linkable":140,"proposed":65,"self":65},{"name":325,"methodId":326,"linkable":140,"proposed":65,"self":65},"LSD-SLAM","lsdslam2014",{"name":328,"methodId":58,"linkable":65,"proposed":65,"self":65},"Submap BA",{"name":219,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":221,"methodId":184,"linkable":140,"proposed":65,"self":65},{"name":223,"methodId":184,"linkable":140,"proposed":65,"self":65},{"name":225,"methodId":184,"linkable":140,"proposed":140,"self":65},[334,336,337,339,340,342,343,345,347,349,350,352,354,356,357,358,360,361,363,364,366,367,368,369,371,372,374,375,376,377,378,379,380,382,384,385,387,389,390,391,392,394,395,397,398,399,400,401],[143,143,143,335,145,143,145,145,143],2.1,[143,143,147,240,145,143,145,145,143],[143,143,150,338,145,143,145,145,143],3.5,[143,143,153,240,145,143,145,145,143],[147,143,143,341,145,143,145,145,143],2.3,[147,143,147,238,145,143,145,145,143],[147,143,150,344,145,143,145,145,143],3.2,[147,143,153,346,145,143,145,145,143],1.7,[150,143,143,348,145,143,145,145,143],4.3,[150,143,147,150,145,143,145,145,143],[150,143,150,351,145,143,145,145,143],4.2,[150,143,153,353,145,143,145,145,143],201.8,[153,143,143,355,145,143,145,145,143],3.7,[153,143,147,230,145,143,145,145,143],[153,143,150,153,145,143,145,145,143],[153,143,153,359,145,143,145,145,143],3.1,[156,143,143,341,145,143,145,145,143],[156,143,147,362,145,143,145,145,143],2.2,[156,143,150,341,145,143,145,145,143],[156,143,153,365,145,143,145,145,143],8.7,[93,143,143,150,145,143,145,145,143],[93,143,147,286,145,143,145,145,143],[93,143,150,346,145,143,145,145,143],[93,143,153,370,145,143,145,145,143],1.6,[161,143,143,58,143,143,145,145,143],[161,143,147,373,145,143,145,145,143],1.5,[161,143,150,58,143,143,145,145,143],[161,143,153,58,143,143,145,145,143],[163,143,143,362,145,143,145,145,143],[163,143,147,58,143,143,145,145,143],[163,143,150,338,145,143,145,145,143],[163,143,153,58,143,143,145,145,143],[148,143,143,381,145,143,145,145,143],2.7,[148,143,147,383,145,143,145,145,143],9.1,[148,143,150,153,145,143,145,145,143],[148,143,153,386,145,143,145,145,143],192.9,[289,143,143,388,145,143,145,145,143],1.9,[289,143,147,259,145,143,145,145,143],[289,143,150,230,145,143,145,145,143],[289,143,153,370,145,143,145,145,143],[393,143,143,346,145,143,145,145,143],10,[393,143,147,259,145,143,145,145,143],[393,143,150,396,145,143,145,145,143],2.8,[393,143,153,259,145,143,145,145,143],[99,143,143,370,145,143,145,145,143],[99,143,147,286,145,143,145,145,143],[99,143,150,362,145,143,145,145,143],[99,143,153,279,145,143,145,145,143],[403],"no value in source ('-')",[303],[],[],[408],"TUM RGB-D ATE RMSE; ground truth from a calibrated motion capture system for hand-held Kinect sequences; for Kinect data the dense reprojection threshold is 0.3 m and residuals above 0.16 m are pruned; Redwood offline and geometry-only",{"slug":410,"group":411,"sourceId":412,"sourceLabel":413,"table":414,"selfRows":153,"metrics":415,"seqs":418,"entrants":423,"cells":442,"outcomes":478,"locators":479,"hardware":480,"wordings":481,"notes":482},"badslam2019-table-2","badslam2019:Table 2","badslam2019","Schöps et al., 2019","Table 2",[416],{"label":417,"unit":190,"statistic":191,"alignment":33},"ATE RMSE [cm]",[419,421,422],{"dataset":308,"sequence":309,"environment":420},"TUM RGB-D real-world sequences recorded with a Kinect v1 (rolling shutter; depth and colour streams not synchronised, Sec. 5); scene type and carrying mode not described in this paper",{"dataset":308,"sequence":312,"environment":420},{"dataset":308,"sequence":314,"environment":420},[424,426,427,429,430,431,434,436,437,438,440],{"name":425,"methodId":184,"linkable":140,"proposed":65,"self":65},"BundleFusion",{"name":205,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":428,"methodId":217,"linkable":140,"proposed":65,"self":65},"ElasticFusion",{"name":211,"methodId":212,"linkable":140,"proposed":65,"self":65},{"name":209,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":432,"methodId":433,"linkable":140,"proposed":65,"self":65},"ORB-SLAM2","orbslam2_2017",{"name":435,"methodId":58,"linkable":65,"proposed":65,"self":65},"PSM SLAM",{"name":207,"methodId":58,"linkable":65,"proposed":65,"self":65},{"name":214,"methodId":5,"linkable":140,"proposed":65,"self":140},{"name":439,"methodId":412,"linkable":140,"proposed":65,"self":65},"BAD SLAM ablation: Ours (fixed intr.)",{"name":441,"methodId":412,"linkable":140,"proposed":140,"self":65},"BAD SLAM (Ours)",[443,444,445,446,447,448,449,450,451,452,453,454,455,456,457,458,459,460,461,462,463,464,465,466,467,468,469,470,472,473,475,476,477],[143,143,143,370,145,143,145,145,143],[143,143,147,286,145,143,145,145,143],[143,143,150,362,145,143,145,145,143],[147,143,143,335,145,143,145,145,143],[147,143,147,240,145,143,145,145,143],[147,143,150,338,145,143,145,145,143],[150,143,143,150,145,143,145,145,143],[150,143,147,286,145,143,145,145,143],[150,143,150,346,145,143,145,145,143],[153,143,143,355,145,143,145,145,143],[153,143,147,230,145,143,145,145,143],[153,143,150,153,145,143,145,145,143],[156,143,143,348,145,143,145,145,143],[156,143,147,150,145,143,145,145,143],[156,143,150,351,145,143,145,145,143],[93,143,143,370,145,143,145,145,143],[93,143,147,261,145,143,145,145,143],[93,143,150,147,145,143,145,145,143],[161,143,143,370,145,143,145,145,143],[161,143,147,58,143,143,145,145,143],[161,143,150,359,145,143,145,145,143],[163,143,143,341,145,143,145,145,143],[163,143,147,238,145,143,145,145,143],[163,143,150,344,145,143,145,145,143],[148,143,143,341,145,143,145,145,143],[148,143,147,362,145,143,145,145,143],[148,143,150,341,145,143,145,145,143],[289,143,143,471,145,143,145,145,143],3.6,[289,143,147,279,145,143,145,145,143],[289,143,150,474,145,143,145,145,143],2.5,[393,143,143,346,145,143,145,145,143],[393,143,147,286,145,143,145,145,143],[393,143,150,346,145,143,145,145,143],[403],[414],[],[],[483],"TUM RGB-D ATE RMSE in cm (rank column omitted); values of other methods copied by the authors from BundleFusion, PSM SLAM and ORB-SLAM2 papers; 'fixed intr.' disables intrinsics and depth-deformation optimisation",[485],{"group":486,"slug":487,"sourceLabel":488,"table":489,"selfRows":150,"datasets":490},"infinitam2015:Table 1","infinitam2015-table-1","Kähler et al., 2015","Table 1",[491,492],"authors' couch sequence","authors' teddy sequence",1790510666147]