[{"data":1,"prerenderedAt":396},["ShallowReactive",2],{"method-yuan2026_adaptive3dgsslam":3},{"method":4,"reference":55,"equipment":77,"figures":99,"results":100},{"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":28,"sensors":36,"platform":39,"estimator":42,"association":43,"timeModel":44,"deskew":45,"loopClosure":46,"globalOptimization":47,"mapRepresentation":48,"prior":49,"outputGeometry":50,"compute":51,"codeUrl":52,"codeLicense":53,"relatedVersions":54},"yuan2026_adaptive3dgsslam","Yuan et al., 2026","Adaptive 3DGS-SLAM (indoor digital twinning)","Adaptive 3DGS-SLAM-driven incremental online geometric digital twinning complex indoor built environments",2026,"recent","C09","odometry_with_local_mapping","本研究針對室內建成環境的線上幾何數位孿生更新，以 MonoGS 預先建立的基準 3DGS 模型為先驗，提出自適應 3DGS-SLAM：新進 RGB-D 影格先以渲染比對方式對齊基準模型求位姿，再以高斯模糊後的滑動視窗 SSIM 產生變化遮罩，變化像素比例超過 2% 且符合原關鍵影格準則者才成為更新用關鍵影格；接著在 CUDA 反向傳播中累計各高斯對變化像素的貢獻，將高貢獻高斯及其鄰域隨機軟剪除一半，並只由遮罩區的 RGB-D 像素加入新高斯。ReplicaCAD 模擬中，相較 MonoGS 直接更新，每影格平均時間由 2.730 秒降至 1.162 秒（降低 57.4%），測試視角 SSIM 由 0.6537 升至 0.7542，ATE RMSE 約 0.105 m；在約 20 平方公尺實驗室以 Intel RealSense D435 實測，每影格時間降低 21.5%，SSIM 由 0.80 升至 0.84。全文未評估點雲或表面幾何精度，也沒有閉環與光束法平差。","RGB-D 3DGS-SLAM that tracks new frames against a prior MonoGS Gaussian model, detects changes with a sliding-window SSIM mask and edits only the Gaussians contributing to changed pixels; 57.4% faster and higher test-view SSIM than MonoGS updating on ReplicaCAD, 21.5% faster in a 20 m2 lab; rendering metrics only, ATE about 0.1 m, no loop closure.","full_text_reviewed","peer_reviewed_published","main_body","營建資訊期刊中處理「既有 3DGS 數位孿生於例行巡檢時增量更新」的研究（作者指出 3DGS-SLAM 的增量更新仍少有探討），對設施管理與室內改修追蹤有參考價值；但驗證僅有 ReplicaCAD 模擬與單一約 20 平方公尺實驗室，評估以渲染指標（PSNR、SSIM）與模擬 ATE 為主，未量測幾何誤差，模擬中的追蹤誤差約 10 公分且無閉環，不宜作為量測等級竣工幾何的證據。4D 施工進度監測只是作者在討論中的展望，並未經實驗驗證。",[20,21],"simulation","controlled_experiment",[23,24,25,26,27],"On ReplicaCAD, test-view SSIM 0.7542 and PSNR 19.48 dB versus 0.6537 and 13.98 dB for MonoGS updating (Table 3)","Average time per frame 1.162 s versus 2.730 s for MonoGS (57.4% lower); keyframe mapping 2.772 s versus 9.345 s (Table 4)","Real lab case: PSNR 23.37 dB, SSIM 0.84, 0.625 s per frame versus 22.66 dB, 0.80, 0.796 s for MonoGS updating; 37% fewer processed frames (Table 5, Sec. 5.4)","Keeps unchanged regions that rebuild baselines (SplaTAM, Photo-SLAM, GS-ICP-SLAM) lose when only the update scan is used (Sec. 4.6, Fig. 8)","Mitigates catastrophic forgetting and visual tearing seen in MonoGS updating (Fig. 12)",[29,30,31,32,33,34,35],"Tracking ATE RMSE about 0.105 m on ReplicaCAD, far worse than SplaTAM (0.0086 m) and Photo-SLAM (0.040 m), because of lightweight frame-to-model alignment without backend optimisation (Table 3, Sec. 4.6)","No loop closure or bundle adjustment (Sec. 6.2)","Assumes static environments; dynamic objects and occlusions not handled (Sec. 6.2)","Real-world validation limited to a single laboratory case of about 20 m2 (Sec. 5.1, Sec. 6.2)","No semantic information (Sec. 6.2)","RGB-only configuration suffers depth ambiguity: ATE RMSE 1.4795 m and 2.321 s per frame (Table 3, Sec. 4.6)","No point or surface geometric accuracy metric is reported anywhere in the full text (Sec. 4 to 5)",[37,38],"RGB-D camera: Intel RealSense D435, 1280 x 720 px at 30 frames per second (real-world case)","simulated RGB-D frames from ReplicaCAD (computer experiment)",[40,41],"simulation (ReplicaCAD FRL apartment, simulated human roaming trajectories)","not_reported (real case: RealSense D435 in an indoor lab; handheld or robot carrier not stated)","frame-to-model tracking inherited from MonoGS: each frame's 6-DoF pose (unit quaternion and translation), initialised from the previous pose, is optimised by minimising a rendering-based loss between the captured image and the baseline 3DGS rendering (Eq. 4: L1 plus SSIM terms; Sec. 4.6 states RGB-D gives joint photometric and geometric supervision); no backend bundle adjustment","dense rendering-based alignment (no feature matching); change detection by sliding-window SSIM between Gaussian-blurred rendered and captured images, pixels below a threshold form a change mask, and a frame becomes an update keyframe only if the masked ratio exceeds 2% and the MonoGS keyframe criteria are also met","discrete frame-by-frame processing of streaming RGB-D frames (30 fps capture in the real case); the design runs the frontend (tracking, change detection) in parallel with the backend (editing, mapping) as in MonoGS (Sec. 3.1), but all reported Python timings were run in single-thread mode (Sec. 4.3)","not_applicable","none (authors state the framework has no loop closure detection)","none (no global or local bundle adjustment); only windowed keyframe 3DGS map optimisation as in MonoGS plus a final colour refinement of spherical harmonic coefficients","3D Gaussians (MonoGS map); each Gaussian's contribution to changed pixels (alpha times transmittance, accumulated in the modified CUDA backward pass) selects candidates; candidates and neighbours within radius r are soft-pruned at random with probability 0.5 (opacity set to 0.01); new Gaussians are added only from masked RGB-D pixels and are protected while their source keyframe is in the window buffer","baseline 3DGS model built beforehand with MonoGS from an earlier RGB-D session (ReplicaCAD Baked_sc0_staging_01; 1225 real frames in the lab case)","updated 3DGS model (about 128,251 Gaussians in the ReplicaCAD run, about 70% kept from the baseline); no point cloud or surface accuracy is evaluated; outputs assessed by rendering metrics (PSNR, SSIM) and ATE RMSE on simulation","Python implementation on MonoGS with a modified CUDA rasteriser backward pass; AMD EPYC 9654 (16 cores in a VM), 60 GB RAM, single Nvidia 4090 GPU, Ubuntu 20.04, CUDA 11.6, single-thread mode; 1.162 s per frame on ReplicaCAD and 0.625 s per frame in the real case",null,"not_verified",[],{"id":5,"kind":56,"shortName":7,"title":8,"authors":57,"year":9,"venue":64,"venueType":65,"publisher":66,"volumeIssuePages":67,"doi":68,"arxivId":52,"url":69,"firstPublicDate":70,"publicationStatus":16,"metadataStatus":71,"fulltextStatus":15,"era":10,"classicReason":45,"codeUrl":52,"cluster":11,"topics":72,"mdpi":73,"verification":74,"label":6,"fulltextRoute":75,"versionRead":76,"addedByCensus":73},"method",[58,59,60,61,62,63],"Ye Yuan","Long Chen","Qiuchen Lu","Thomas Shiu Tong Ng","Hongyang Li","Shanjing Zhou","Advanced Engineering Informatics","journal","Elsevier","76, 105077","10.1016\u002Fj.aei.2026.105077","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1016\u002Fj.aei.2026.105077","2026-07-24","metadata_verified",[11],false,"corrected","publisher OA","version of record, Advanced Engineering Informatics 76 Part C (November 2026) 105077, ScienceDirect HTML full text (open access, CC BY-NC-ND 4.0)",[78,84,90,96],{"category":79,"model":80,"canonical":80,"role":81,"dataset":52,"specs":82,"locator":83},"rgbd","Intel RealSense D435","method input","RGB-D images at 30 frames per second, 1280 x 720 px; 1652 frames (1225 for the baseline model, 427 for updating)","Sec. 5.1",{"category":79,"model":85,"canonical":85,"role":86,"dataset":87,"specs":88,"locator":89},"simulated RGB-D camera (human roaming trajectories; sensor not named)","dataset sensor","ReplicaCAD (FRL apartment, Baked_sc0_staging_01 and Baked_sc0_staging_05)","RGB-D sequences organised in the Replica dataset format; different paths for the baseline and update configurations; intrinsics not reported","Sec. 4.1",{"category":91,"model":92,"canonical":92,"role":93,"dataset":52,"specs":94,"locator":95},"compute","AMD EPYC 9654 96-Core Processor","compute for runtime","16 cores allocated within a virtual machine, 60 GB memory, Ubuntu 20.04","Sec. 4.3, Sec. 5.2",{"category":91,"model":97,"canonical":97,"role":93,"dataset":52,"specs":98,"locator":95},"Nvidia 4090 GPU","single GPU, CUDA 11.6",[],{"totalRows":101,"groupCount":102,"groups":103,"others":385},43,6,[104,208,284,338],{"slug":105,"group":106,"sourceId":5,"sourceLabel":6,"table":107,"selfRows":108,"metrics":109,"seqs":130,"entrants":135,"cells":145,"outcomes":201,"locators":202,"hardware":203,"wordings":205,"notes":206},"yuan2026-adaptive3dgsslam-table-2","yuan2026_adaptive3dgsslam:Table 2","Table 2",24,[110,115,117,120,122,126],{"label":111,"unit":112,"statistic":113,"alignment":114},"Training-view SSIM","unitless","not_reported","none",{"label":116,"unit":112,"statistic":113,"alignment":114},"Test-view SSIM",{"label":118,"unit":119,"statistic":113,"alignment":114},"Training-view PSNR","dB",{"label":121,"unit":119,"statistic":113,"alignment":114},"Test-view PSNR",{"label":123,"unit":124,"statistic":125,"alignment":113},"ATE RMSE [m]","m","RMSE",{"label":127,"unit":128,"statistic":129,"alignment":114},"Avg. time [s]","s","mean",[131],{"dataset":132,"sequence":133,"environment":134},"ReplicaCAD (FRL apartment)","update Baked_sc0_staging_05 (baseline from Baked_sc0_staging_01)","simulated indoor apartment with furniture rearrangement",[136,139,141,143],{"name":137,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method, change ratio threshold >=1%",true,{"name":140,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method, change ratio threshold >=2% (default)",{"name":142,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method, change ratio threshold >=3.5%",{"name":144,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method, change ratio threshold >=5%",[146,150,153,156,159,162,165,167,169,171,173,175,177,179,181,183,185,187,189,191,193,195,197,199],[147,147,147,148,149,147,149,149,147],0,0.8295,-1,[147,151,147,152,149,147,149,149,147],1,0.7304,[147,154,147,155,149,147,149,149,147],2,22.51,[147,157,147,158,149,147,149,149,147],3,18.54,[147,160,147,161,149,147,149,149,147],4,0.0849,[147,163,147,164,149,147,147,149,147],5,1.181,[151,147,147,166,149,147,149,149,147],0.8354,[151,151,147,168,149,147,149,149,147],0.7542,[151,154,147,170,149,147,149,149,147],22.75,[151,157,147,172,149,147,149,149,147],19.48,[151,160,147,174,149,147,149,149,147],0.1054,[151,163,147,176,149,147,147,149,147],1.162,[154,147,147,178,149,147,149,149,147],0.8329,[154,151,147,180,149,147,149,149,147],0.7377,[154,154,147,182,149,147,149,149,147],22.74,[154,157,147,184,149,147,149,149,147],18.77,[154,160,147,186,149,147,149,149,147],0.0862,[154,163,147,188,149,147,147,149,147],1.003,[157,147,147,190,149,147,149,149,147],0.824,[157,151,147,192,149,147,149,149,147],0.7369,[157,154,147,194,149,147,149,149,147],21.81,[157,157,147,196,149,147,149,149,147],19.05,[157,160,147,198,149,147,149,149,147],0.1068,[157,163,147,200,149,147,147,149,147],1.023,[],[107],[204],"AMD EPYC 9654 (16 cores in a VM), 60 GB RAM, single Nvidia 4090 GPU, Ubuntu 20.04, CUDA 11.6; single-thread Python",[],[207],"Change-ratio threshold sensitivity of the proposed method on ReplicaCAD; training-view metrics on update frames, test-view metrics on 100 separately generated poses; Avg. time = total time \u002F scanned frames",{"slug":209,"group":210,"sourceId":5,"sourceLabel":6,"table":211,"selfRows":212,"metrics":213,"seqs":218,"entrants":221,"cells":237,"outcomes":278,"locators":279,"hardware":280,"wordings":281,"notes":282},"yuan2026-adaptive3dgsslam-table-3","yuan2026_adaptive3dgsslam:Table 3","Table 3",8,[214,215,216,217],{"label":116,"unit":112,"statistic":113,"alignment":114},{"label":121,"unit":119,"statistic":113,"alignment":114},{"label":123,"unit":124,"statistic":125,"alignment":113},{"label":127,"unit":128,"statistic":129,"alignment":114},[219],{"dataset":132,"sequence":220,"environment":134},"update Baked_sc0_staging_05 (1000 frames)",[222,224,227,230,233,235],{"name":223,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method",{"name":225,"methodId":226,"linkable":138,"proposed":73,"self":73},"Update with MonoGS","monogs2024",{"name":228,"methodId":229,"linkable":138,"proposed":73,"self":73},"Rebuild with SplaTAM","splatam2024",{"name":231,"methodId":232,"linkable":138,"proposed":73,"self":73},"Rebuild with Photo-SLAM","photoslam2024",{"name":234,"methodId":52,"linkable":73,"proposed":73,"self":73},"Rebuild with GS-ICP-SLAM",{"name":236,"methodId":5,"linkable":138,"proposed":138,"self":138},"Proposed method (RGB-only)",[238,239,240,241,242,244,246,248,250,252,254,256,258,260,262,264,266,268,270,272,274,276],[147,147,147,168,149,147,149,149,147],[147,151,147,172,149,147,149,149,147],[147,154,147,174,149,147,149,149,147],[147,157,147,176,149,147,147,149,147],[151,147,147,243,149,147,149,149,147],0.6537,[151,151,147,245,149,147,149,149,147],13.98,[151,154,147,247,149,147,149,149,147],0.1109,[151,157,147,249,149,147,147,149,147],2.73,[154,147,147,251,149,147,149,149,147],0.6834,[154,151,147,253,149,147,149,149,147],13.39,[154,154,147,255,149,147,149,149,147],0.0086,[154,157,147,257,149,147,147,149,147],5.988,[157,147,147,259,149,147,149,149,147],0.7339,[157,151,147,261,149,147,149,149,147],15.14,[157,154,147,263,149,147,149,149,147],0.04,[160,147,147,265,149,147,149,149,147],0.581,[160,151,147,267,149,147,149,149,147],9.04,[160,154,147,269,149,147,149,149,147],1.0764,[163,147,147,271,149,147,149,149,147],0.6696,[163,151,147,273,149,147,149,149,147],13.58,[163,154,147,275,149,147,149,149,147],1.4795,[163,157,147,277,149,147,147,149,147],2.321,[],[211],[204],[],[283],"ReplicaCAD comparison: proposed and MonoGS update the baseline model, SplaTAM, Photo-SLAM and GS-ICP-SLAM rebuild from the update sequence only; test views = 100 generated poses; Photo-SLAM and GS-ICP-SLAM times from default multi-process runs, others single-thread Python",{"slug":285,"group":286,"sourceId":5,"sourceLabel":6,"table":287,"selfRows":102,"metrics":288,"seqs":302,"entrants":304,"cells":308,"outcomes":332,"locators":333,"hardware":334,"wordings":335,"notes":336},"yuan2026-adaptive3dgsslam-table-4","yuan2026_adaptive3dgsslam:Table 4","Table 4",[289,291,293,296,298,300],{"label":290,"unit":128,"statistic":129,"alignment":114},"Tracking [s] (all frames)",{"label":292,"unit":128,"statistic":129,"alignment":114},"Keyframe determine [s] (all frames)",{"label":294,"unit":295,"statistic":113,"alignment":114},"Num. keyframes","count",{"label":297,"unit":128,"statistic":129,"alignment":114},"Gaussian edit [s] (keyframes)",{"label":299,"unit":128,"statistic":129,"alignment":114},"Mapping [s] (keyframes)",{"label":301,"unit":128,"statistic":129,"alignment":114},"Whole process Avg. time [s]",[303],{"dataset":132,"sequence":220,"environment":134},[305,306,307],{"name":223,"methodId":5,"linkable":138,"proposed":138,"self":138},{"name":225,"methodId":226,"linkable":138,"proposed":73,"self":73},{"name":228,"methodId":229,"linkable":138,"proposed":73,"self":73},[309,311,313,315,317,319,320,322,324,326,328,330,331],[147,147,147,310,149,147,147,149,147],0.66,[147,151,147,312,149,147,147,149,147],0.016,[147,154,147,314,149,147,149,149,147],172,[147,157,147,316,149,147,147,149,147],0.017,[147,160,147,318,149,147,147,149,147],2.772,[147,163,147,176,149,147,147,149,147],[151,147,147,321,149,147,147,149,147],0.618,[151,151,147,323,149,147,147,149,147],0.01,[151,154,147,325,149,147,149,149,147],224,[151,157,147,327,149,147,147,149,147],0.043,[151,160,147,329,149,147,147,149,147],9.345,[151,163,147,249,149,147,147,149,147],[154,163,147,257,149,147,147,149,147],[],[287],[204],[],[337],"Per-step runtime on the ReplicaCAD update sequence (1000 frames); tracking and keyframe determination averaged over all frames, Gaussian edit and mapping over keyframes only; whole-process average = total time \u002F scanned frames; single-thread Python",{"slug":339,"group":340,"sourceId":5,"sourceLabel":6,"table":341,"selfRows":157,"metrics":342,"seqs":348,"entrants":353,"cells":360,"outcomes":377,"locators":378,"hardware":380,"wordings":382,"notes":383},"yuan2026-adaptive3dgsslam-table-5","yuan2026_adaptive3dgsslam:Table 5","Table 5",[343,345,347],{"label":344,"unit":119,"statistic":129,"alignment":114},"PSNR [dB]",{"label":346,"unit":112,"statistic":129,"alignment":114},"SSIM",{"label":127,"unit":128,"statistic":129,"alignment":114},[349],{"dataset":350,"sequence":351,"environment":352},"authors' real-world RGB-D dataset","PSNR and SSIM: 300 random frames evenly sampled from the initial (1225) and updating (427) subsets; Avg. time: total runtime divided by the number of frames in the dataset (Sec. 5.4)","indoor laboratory of about 20 m2 with shelves, pipelines, equipment and computers",[354,356,358],{"name":355,"methodId":5,"linkable":138,"proposed":138,"self":138},"Adaptive 3DGS-SLAM",{"name":357,"methodId":226,"linkable":138,"proposed":73,"self":73},"MonoGS updating pipeline",{"name":359,"methodId":226,"linkable":138,"proposed":73,"self":73},"Baseline without updating (MonoGS initial model)",[361,363,365,367,369,371,373,375],[147,147,147,362,149,147,149,149,147],23.37,[147,151,147,364,149,147,149,149,147],0.84,[147,154,147,366,149,151,147,149,147],0.625,[151,147,147,368,149,147,149,149,147],22.66,[151,151,147,370,149,147,149,149,147],0.8,[151,154,147,372,149,151,147,149,147],0.796,[154,147,147,374,149,147,149,149,147],21.27,[154,151,147,376,149,147,149,149,147],0.83,[],[341,379],"Table 5, Sec. 5.2",[381],"same platform as Sec. 4 (AMD EPYC 9654, 16 cores in a VM, 60 GB RAM, single Nvidia 4090 GPU)",[],[384],"Real-world lab case (Intel RealSense D435; 1225 baseline frames then 427 update frames); PSNR and SSIM are means over 300 random frames sampled from both subsets; Avg. time = total runtime \u002F all frames",[386,391],{"group":387,"slug":388,"sourceLabel":6,"table":389,"selfRows":151,"datasets":390},"yuan2026_adaptive3dgsslam:Text Sec.4.4","yuan2026-adaptive3dgsslam-text-sec-4-4","Text Sec.4.4",[132],{"group":392,"slug":393,"sourceLabel":6,"table":394,"selfRows":151,"datasets":395},"yuan2026_adaptive3dgsslam:Text Sec.5.4","yuan2026-adaptive3dgsslam-text-sec-5-4","Text Sec.5.4",[350],1790510663841]