[{"data":1,"prerenderedAt":427},["ShallowReactive",2],{"method-choi2015robustrecon":3},{"method":4,"reference":57,"equipment":79,"figures":86,"results":87},{"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":22,"limitations":27,"sensors":32,"platform":34,"estimator":37,"association":38,"timeModel":39,"deskew":40,"loopClosure":41,"globalOptimization":42,"mapRepresentation":43,"prior":44,"outputGeometry":45,"compute":46,"codeUrl":47,"codeLicense":48,"relatedVersions":49},"choi2015robustrecon","Choi et al., 2015","Robust Reconstruction of Indoor Scenes (Redwood)","Robust reconstruction of indoor scenes",2015,"classic","C08","offline_map_refinement","本文提出離線的 RGB-D 室內場景重建流程。先把影片切成每 50 張影格一段，以 RGB-D 里程計估計段內軌跡並以 TSDF 融合成場景片段（fragment）；再對所有片段兩兩做幾何全域配準（改良的 PCL FPFH 搭配 RANSAC），得到候選迴圈閉合。作者發現即使最佳的配準演算法，精確率也低於 20%，因此在位姿圖最佳化中為每條迴圈閉合邊加入線過程（line process）變數，並以稠密表面對應距離定義對齊誤差，讓錯誤的邊在同一個最小平方問題中自動失效。修剪後再以 ICP 精修與位姿圖求得片段位姿，最後以體積整合輸出全域網格。作者同時擴充 ICL-NUIM 資料集，加入完整掃描軌跡、較真實的深度雜訊模型與辦公室場景表面真值。","Offline fragment-based RGB-D reconstruction: 50-frame TSDF fragments are registered pairwise by geometric global registration, and a pose graph with line-process variables on loop-closure edges and a dense surface-alignment cost prunes false loop closures before ICP refinement and volumetric integration; also introduces the augmented ICL-NUIM benchmark (full-scan trajectories, realistic noise, office ground truth).","full_text_reviewed","peer_reviewed_published","background","論文未在施工現場測試，真實場景評估為 SUN3D 的住宅、旅館與校園室內，且以群眾外包成對比較評分，並非對照量測幾何真值。以片段為單位的離線重建與穩健迴圈閉合，適合工地資料事後處理的整層室內重建；Fig. 1 的公寓重建軌跡長 151.6 公尺，但單一序列需數小時處理。corpus 中 bundlefusion2017 以 Redwood 作為離線基準。",[20,21],"simulation","public_benchmark",[23,24,25,26],"Loop-closure precision rises from 19.6% to 97.7% while recall drops only from 59.2% to 57.8% on augmented ICL-NUIM (Table 2)","Mean distance to the ground-truth surface 0.03 to 0.07 m (average 0.05 m), about half that of the next best pipeline, SUN3D SfM (Table 4)","Finds loop closures from geometry even when the corresponding images are not similar (Fig. 2)","Ranked above all automatic pipelines on eight SUN3D scenes by crowd-sourced comparisons, and above the manually assisted models on 6 of 8 (Table 7)",[28,29,30,31],"Offline: 29 to 187 minutes per evaluated sequence and 387 minutes for the 17,391-frame apartment on a desktop CPU (supplementary Table 1)","If the input contains no loop closures, odometry drift accumulates and distorts the model (Sec. 7)","Catastrophic odometry failure inside a fragment is not handled (Sec. 7)","Real-world scenes are evaluated only perceptually by crowd-sourced pairwise comparisons, not against measured geometry (Sec. 6.3)",[33],"RGB-D video from a consumer depth camera (SUN3D real scenes; augmented ICL-NUIM synthetic sequences with a disparity-based noise and distortion model); fragment-pair registration is geometric (FPFH on fragment point sets), sensor models not named",[35,36],"real indoor RGB-D sequences (SUN3D scenes and an apartment sequence; capture platform not described in the paper)","simulation (augmented ICL-NUIM living room and office; trajectories model thorough handheld scanning)","Offline pipeline: RGB-D odometry (Kerl et al. 2013) inside 50-frame fragments; pose graph over fragment poses with odometry edges and loop-closure edges weighted by line-process variables l in [0, 1] with prior (sqrt(l) - 1)^2 and balance mu = tau^2 kappa, solved with g2o; edges with l \u003C 0.25 pruned; ICP refinement of the remaining edges and final pose-graph optimization; optional SLAC non-rigid refinement","Pairwise geometric registration of every fragment pair with a modified PCL FPFH plus RANSAC algorithm (four-point samples, normal-angle, edge-length and overlap checks); candidates kept when more than 30% overlap; alignment cost uses dense point correspondences within 5 cm","discrete poses (fragment poses; frame poses within fragments from odometry)","not_applicable (RGB-D input)","Geometric: every fragment pair tested by global registration, so loops are found even when images are not similar; false candidates removed by the line-process optimization","Offline robust pose-graph optimization of fragments (line processes, g2o), then ICP refinement and pose-graph optimization; optional non-rigid refinement","fragment meshes from volumetric TSDF integration (Curless and Levoy) of 50-frame segments; final global volumetric integration","none","global surface mesh (for example 15.8 million triangles for the 17,391-frame apartment)","Offline on a workstation with Intel Core i7-3770 3.5 GHz CPU and 16 GB RAM; total 29 to 187 minutes per evaluated sequence and 387 minutes for the apartment (supplementary Table 1)","https:\u002F\u002Fgithub.com\u002Fqianyizh\u002FElasticReconstruction","MIT-style licence (LICENSE.txt checked, Copyright 2013 Stanford University)",[50,53],{"relation":51,"title":52,"doi_or_url":47},"code_release","qianyizh\u002FElasticReconstruction",{"relation":54,"title":55,"doi_or_url":56},"dataset","Augmented ICL-NUIM dataset and indoor reconstruction pages (redwood-data.org\u002Findoor)","http:\u002F\u002Fredwood-data.org\u002Findoor\u002F",{"id":5,"kind":58,"shortName":7,"title":8,"authors":59,"year":9,"venue":63,"venueType":64,"publisher":65,"volumeIssuePages":66,"doi":67,"arxivId":68,"url":69,"firstPublicDate":70,"publicationStatus":16,"metadataStatus":71,"fulltextStatus":15,"era":10,"classicReason":72,"codeUrl":47,"cluster":11,"topics":73,"mdpi":74,"verification":75,"label":6,"fulltextRoute":76,"versionRead":77,"addedByCensus":78},"method",[60,61,62],"Sungjoon Choi","Qian-Yi Zhou","Vladlen Koltun","2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)","conference","IEEE","pp. 5556-5565","10.1109\u002Fcvpr.2015.7299195",null,"https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1109\u002FCVPR.2015.7299195","2015-06-07","metadata_verified","principle reused: offline fragment-based RGB-D reconstruction in which false loop closures from geometric fragment registration are pruned inside the pose-graph optimization by line-process variables with a dense surface-alignment cost; also introduced the augmented ICL-NUIM evaluation (full-scan trajectories, realistic noise, office ground truth). Appears as the offline 'Redwood' baseline in bundlefusion2017.",[11],false,"corrected","other","CVF open-access version of the CVPR 2015 paper (identical content to the conference paper per CVF; IEEE Xplore version of record not compared) plus the authors' supplementary PDF from the Redwood project page",true,[80],{"category":81,"model":82,"canonical":82,"role":83,"dataset":68,"specs":84,"locator":85},"compute","workstation with Intel Core i7-3770 3.5GHz CPU and 16GB of RAM","compute for runtime","all pipeline steps; registration timings single-threaded","Table 1 caption; supplementary Table 1",[],{"totalRows":88,"groupCount":89,"groups":90,"others":411},53,7,[91,163,243,354],{"slug":92,"group":93,"sourceId":5,"sourceLabel":6,"table":94,"selfRows":95,"metrics":96,"seqs":103,"entrants":114,"cells":119,"outcomes":156,"locators":157,"hardware":158,"wordings":159,"notes":160},"choi2015robustrecon-table-2","choi2015robustrecon:Table 2","Table 2",16,[97,101],{"label":98,"unit":99,"statistic":100,"alignment":44},"Recall (%)","%","not_reported",{"label":102,"unit":99,"statistic":100,"alignment":44},"Precision (%)",[104,108,110,112],{"dataset":105,"sequence":106,"environment":107},"augmented ICL-NUIM (synthetic, realistic noise, full-scan trajectories)","Living room 1","synthetic living room and office",{"dataset":105,"sequence":109,"environment":107},"Living room 2",{"dataset":105,"sequence":111,"environment":107},"Office 1",{"dataset":105,"sequence":113,"environment":107},"Office 2",[115,117],{"name":116,"methodId":5,"linkable":78,"proposed":74,"self":78},"Geometric registration candidates before pruning",{"name":118,"methodId":5,"linkable":78,"proposed":78,"self":78},"After line-process pruning (Ours)",[120,124,127,129,130,133,135,138,140,142,144,146,148,150,152,154],[121,121,121,122,123,121,123,123,121],0,61.2,-1,[125,121,121,126,123,121,123,123,121],1,57.6,[121,121,125,128,123,121,123,123,121],49.7,[125,121,125,128,123,121,123,123,121],[121,121,131,132,123,121,123,123,121],2,64.4,[125,121,131,134,123,121,123,123,121],63.3,[121,121,136,137,123,121,123,123,121],3,61.5,[125,121,136,139,123,121,123,123,121],60.7,[121,125,121,141,123,121,123,123,125],27.2,[125,125,121,143,123,121,123,123,125],95.1,[121,125,125,145,123,121,123,123,125],17,[125,125,125,147,123,121,123,123,125],97.4,[121,125,131,149,123,121,123,123,125],19.2,[125,125,131,151,123,121,123,123,125],98.3,[121,125,136,153,123,121,123,123,125],14.9,[125,125,136,155,123,121,123,123,125],100,[],[94],[],[],[161,162],"Loop-closure recall of fragment pairs (ground-truth loop = more than 30% overlap; true positive if ground-truth correspondence RMSE below 0.2 m)","Loop-closure precision of fragment pairs (ground-truth loop = more than 30% overlap; true positive if ground-truth correspondence RMSE below 0.2 m)",{"slug":164,"group":165,"sourceId":5,"sourceLabel":6,"table":166,"selfRows":167,"metrics":168,"seqs":172,"entrants":198,"cells":201,"outcomes":235,"locators":236,"hardware":238,"wordings":240,"notes":241},"choi2015robustrecon-supp-table-1","choi2015robustrecon:Supp. Table 1","Supp. Table 1",13,[169],{"label":170,"unit":171,"statistic":100,"alignment":44},"Total time (minutes)","min",[173,177,178,179,180,181,184,186,188,190,192,194,196],{"dataset":174,"sequence":175,"environment":176},"authors' apartment sequence","Apartment (17,391 frames)","indoor",{"dataset":105,"sequence":106,"environment":176},{"dataset":105,"sequence":109,"environment":176},{"dataset":105,"sequence":111,"environment":176},{"dataset":105,"sequence":113,"environment":176},{"dataset":182,"sequence":183,"environment":176},"SUN3D","hotel umd",{"dataset":182,"sequence":185,"environment":176},"harvard c5",{"dataset":182,"sequence":187,"environment":176},"harvard c6",{"dataset":182,"sequence":189,"environment":176},"harvard c8",{"dataset":182,"sequence":191,"environment":176},"mit 32 d507",{"dataset":182,"sequence":193,"environment":176},"mit 76 studyroom",{"dataset":182,"sequence":195,"environment":176},"mit dorm next sj",{"dataset":182,"sequence":197,"environment":176},"mit lab hj",[199],{"name":200,"methodId":5,"linkable":78,"proposed":78,"self":78},"Ours",[202,204,206,208,210,213,216,219,221,224,227,230,233],[121,121,121,203,123,121,121,123,121],387,[121,121,125,205,123,121,121,123,121],178,[121,121,131,207,123,121,121,123,121],64,[121,121,136,209,123,121,121,123,121],86,[121,121,211,212,123,121,121,123,121],4,81,[121,121,214,215,123,121,121,123,121],5,33,[121,121,217,218,123,121,121,123,121],6,75,[121,121,89,220,123,121,121,123,121],29,[121,121,222,223,123,121,121,123,121],8,36,[121,121,225,226,123,121,121,123,121],9,187,[121,121,228,229,123,121,121,123,121],10,114,[121,121,231,232,123,121,121,123,121],11,62,[121,121,234,223,123,121,121,123,121],12,[],[237],"Supplementary Table 1",[239],"Intel Core i7-3770 3.5 GHz, 16 GB RAM",[],[242],"Total running time of all pipeline steps (fragment creation, geometric registration, robust optimization, ICP refinement, integration) per sequence",{"slug":244,"group":245,"sourceId":5,"sourceLabel":6,"table":246,"selfRows":222,"metrics":247,"seqs":251,"entrants":261,"cells":272,"outcomes":348,"locators":349,"hardware":350,"wordings":351,"notes":352},"choi2015robustrecon-table-7","choi2015robustrecon:Table 7","Table 7",[248],{"label":249,"unit":250,"statistic":100,"alignment":44},"BRE score","score (-1 to 1)",[252,254,255,256,257,258,259,260],{"dataset":182,"sequence":183,"environment":253},"real indoor scenes (hotel, campus rooms, dormitory, lab)",{"dataset":182,"sequence":185,"environment":253},{"dataset":182,"sequence":187,"environment":253},{"dataset":182,"sequence":189,"environment":253},{"dataset":182,"sequence":191,"environment":253},{"dataset":182,"sequence":193,"environment":253},{"dataset":182,"sequence":195,"environment":253},{"dataset":182,"sequence":197,"environment":253},[262,264,267,269,270],{"name":263,"methodId":68,"linkable":74,"proposed":74,"self":74},"DVO SLAM [34]",{"name":265,"methodId":266,"linkable":78,"proposed":74,"self":74},"Kintinuous [61]","kintinuous2015",{"name":268,"methodId":68,"linkable":74,"proposed":74,"self":74},"SUN3D SfM [65]",{"name":200,"methodId":5,"linkable":78,"proposed":78,"self":78},{"name":271,"methodId":68,"linkable":74,"proposed":74,"self":74},"SUN3D manual (manually assisted reconstructions)",[273,275,277,279,281,283,285,287,289,291,293,295,297,298,300,302,304,305,307,309,311,313,315,316,318,320,322,324,326,328,330,332,334,336,338,339,341,343,345,347],[121,121,121,274,123,121,123,123,121],-0.61,[125,121,121,276,123,121,123,123,121],-0.45,[131,121,121,278,123,121,123,123,121],-0.02,[136,121,121,280,123,121,123,123,121],0.66,[211,121,121,282,123,121,123,123,121],0.56,[121,121,125,284,123,121,123,123,121],-0.49,[125,121,125,286,123,121,123,123,121],-0.01,[131,121,125,288,123,121,123,123,121],-0.65,[136,121,125,290,123,121,123,123,121],0.94,[211,121,125,292,123,121,123,123,121],0.11,[121,121,131,294,123,121,123,123,121],-0.97,[125,121,131,296,123,121,123,123,121],0.05,[131,121,131,286,123,121,123,123,121],[136,121,131,299,123,121,123,123,121],0.96,[211,121,131,301,123,121,123,123,121],-0.15,[121,121,136,303,123,121,123,123,121],-0.7,[125,121,136,274,123,121,123,123,121],[131,121,136,306,123,121,123,123,121],0.39,[136,121,136,308,123,121,123,123,121],0.65,[211,121,136,310,123,121,123,123,121],0.46,[121,121,211,312,123,121,123,123,121],-0.78,[125,121,211,314,123,121,123,123,121],-0.28,[131,121,211,278,123,121,123,123,121],[136,121,211,317,123,121,123,123,121],0.74,[211,121,211,319,123,121,123,123,121],0.36,[121,121,214,321,123,121,123,123,121],-0.52,[125,121,214,323,123,121,123,123,121],-0.47,[131,121,214,325,123,121,123,123,121],0.35,[136,121,214,327,123,121,123,123,121],0.5,[211,121,214,329,123,121,123,123,121],0.19,[121,121,217,331,123,121,123,123,121],-0.26,[125,121,217,333,123,121,123,123,121],-0.2,[131,121,217,335,123,121,123,123,121],-0.23,[136,121,217,337,123,121,123,123,121],0.1,[211,121,217,308,123,121,123,123,121],[121,121,89,340,123,121,123,123,121],-0.12,[125,121,89,342,123,121,123,123,121],-0.57,[131,121,89,344,123,121,123,123,121],0.03,[136,121,89,346,123,121,123,123,121],0.22,[211,121,89,327,123,121,123,123,121],[],[246],[],[],[353],"Perceptual evaluation on real SUN3D scenes: Balanced Rank Estimation scores from 17,640 crowd-sourced pairwise comparisons (Amazon Mechanical Turk), range -1 to 1, higher is better; not a geometric measurement",{"slug":355,"group":356,"sourceId":5,"sourceLabel":6,"table":357,"selfRows":211,"metrics":358,"seqs":363,"entrants":368,"cells":375,"outcomes":404,"locators":405,"hardware":407,"wordings":408,"notes":409},"choi2015robustrecon-supp-table-3","choi2015robustrecon:Supp. Table 3","Supp. Table 3",[359],{"label":360,"unit":361,"statistic":362,"alignment":100},"median distance to ground-truth surface (m)","m","median",[364,365,366,367],{"dataset":105,"sequence":106,"environment":107},{"dataset":105,"sequence":109,"environment":107},{"dataset":105,"sequence":111,"environment":107},{"dataset":105,"sequence":113,"environment":107},[369,370,371,372,373],{"name":265,"methodId":266,"linkable":78,"proposed":74,"self":74},{"name":263,"methodId":68,"linkable":74,"proposed":74,"self":74},{"name":268,"methodId":68,"linkable":74,"proposed":74,"self":74},{"name":200,"methodId":5,"linkable":78,"proposed":78,"self":78},{"name":374,"methodId":68,"linkable":74,"proposed":74,"self":74},"GT trajectory (input depth fused along ground truth, reference)",[376,378,380,382,383,384,385,386,388,389,391,392,393,394,395,397,399,401,402,403],[121,121,121,377,123,121,123,123,121],0.17,[125,121,121,379,123,121,123,123,121],0.16,[131,121,121,381,123,121,123,123,121],0.08,[136,121,121,344,123,121,123,123,121],[211,121,121,344,123,121,123,123,121],[121,121,125,337,123,121,123,123,121],[125,121,125,296,123,121,123,123,121],[131,121,125,387,123,121,123,123,121],0.06,[136,121,125,296,123,121,123,123,121],[211,121,125,390,123,121,123,123,121],0.02,[121,121,131,337,123,121,123,123,121],[125,121,131,381,123,121,123,123,121],[131,121,131,292,123,121,123,123,121],[136,121,131,390,123,121,123,123,121],[211,121,131,396,123,121,123,123,121],0.01,[121,121,136,398,123,121,123,123,121],0.09,[125,121,136,400,123,121,123,123,121],0.07,[131,121,136,387,123,121,123,123,121],[136,121,136,344,123,121,123,123,121],[211,121,136,390,123,121,123,123,121],[],[406],"Supplementary Table 3",[],[],[410],"Median distance of each reconstructed model to the ground-truth surface (supplementary Appendix E)",[412,417,422],{"group":413,"slug":414,"sourceLabel":6,"table":415,"selfRows":211,"datasets":416},"choi2015robustrecon:Supp. Table 4","choi2015robustrecon-supp-table-4","Supp. Table 4",[105],{"group":418,"slug":419,"sourceLabel":6,"table":420,"selfRows":211,"datasets":421},"choi2015robustrecon:Table 4","choi2015robustrecon-table-4","Table 4",[105],{"group":423,"slug":424,"sourceLabel":6,"table":425,"selfRows":211,"datasets":426},"choi2015robustrecon:Table 5","choi2015robustrecon-table-5","Table 5",[105],1790510664335]