[{"data":1,"prerenderedAt":542},["ShallowReactive",2],{"method-pwclonet2021":3},{"method":4,"reference":59,"equipment":82,"figures":98,"results":99},{"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":26,"sensors":32,"platform":34,"estimator":36,"association":37,"timeModel":38,"deskew":39,"loopClosure":40,"globalOptimization":40,"mapRepresentation":41,"prior":42,"outputGeometry":43,"compute":44,"codeUrl":45,"codeLicense":46,"relatedVersions":47},"pwclonet2021","Wang et al., 2021c","PWCLO-Net","PWCLO-Net: Deep LiDAR Odometry in 3D Point Clouds Using Hierarchical Embedding Mask Optimization",2021,"recent","C04","odometry","PWCLO-Net 是直接以原始 3D 點雲學習的監督式 LiDAR 里程計。它借用光流網路的金字塔、變形與代價體（PWC）結構：兩幀點雲先經共享權重的點特徵金字塔，再以注意力代價體建立軟對應；可訓練的嵌入遮罩為每個點加權，以回歸整體位姿並壓低動態物體與雜草等不可靠點的影響。估得的位姿用來變形前一幀，再由粗到細逐層修正位姿與遮罩。它只做逐幀里程計，沒有地圖或迴圈閉合。","Supervised deep LiDAR odometry on raw 3D points using a pyramid, warping and cost-volume (PWC) structure: an attentive point cost volume gives soft correspondences, a trainable embedding mask weights points to regress the pose and suppress outliers, and pose warp-refinement refines the estimate coarse to fine; frame-to-frame only.","full_text_reviewed","peer_reviewed_published","supplementary","與施工沒有直接證據：只在 KITTI 道路資料上以監督方式訓練與測試，且輸入裁成車輛周圍 30 m 見方並移除地面。學習式里程計在工地這種訓練資料缺乏、場景快速變化的環境是否能泛化，論文沒有驗證。它在 NeRF-LOAM 的比較表中被列為基準 [nerfloam2023]。",[20],"public_benchmark",[22,23,24,25],"Mean errors on KITTI 07-10 (trained on 00-06) of 1.085% and 0.490 deg\u002F100m, lower than LO-Net (1.748%, 0.793) and the full LOAM values from [10] (1.145%, 0.498); DMLO has a lower translational mean (1.008%) but a higher rotational mean (0.538) (Table 1)","Better than LodoNet, DeepPCO and the unsupervised method of Cho et al. under each of their training splits (Tables 2-4)","Embedding mask gives low weight to moving cars, cyclists, bushes and weeds without a separate mask network (Sec. 5.3; Fig. 8)","Ablations show large gains from the cost volume and pose warp-refinement (Table 5)",[27,28,29,30,31],"Odometry only; combining the mask with mapping optimization is left to future work (Sec. 6)","Trained and tested only on KITTI; generalization to other sensors or environments is not evaluated (inference from Sec. 4-5)","Inputs are cropped to 30 m x 30 m and ground points below 0.55 m are removed (Sec. 4.1)","Most baseline numbers are copied from LO-Net [10]; only LOAM without mapping was rerun (Table 1 caption)","Inference runtime not reported in the main paper (Sec. 4.2)",[33],"3D LiDAR point coordinates only (KITTI Velodyne; reflectance not used) (Sec. 4.1)",[35],"vehicle (KITTI)","supervised end-to-end network: siamese point feature pyramid (set conv with farthest point sampling and kNN), attentive point cost volume, trainable embedding mask, and three pose warp-refinement modules that refine a quaternion and translation over four levels; multi-level supervised loss with learnable weighting (Sec. 3; Sec. 4.2)","soft correspondences from an attentive cost volume between two consecutive frames (8192 randomly sampled points each) instead of explicit point matching; the embedding mask down-weights dynamic and irregular points (Sec. 3.2; Sec. 3.3; Sec. 4.2; Sec. 5.3)","discrete frame-to-frame relative poses","not described; KITTI clouds used as provided","none","none (frame-to-frame odometry without a map)","requires ground-truth poses for supervised training (main setting trained on KITTI 00-06); inputs cropped to a 30 m x 30 m square and ground below 0.55 m removed (Sec. 4.1)","relative 6-DoF pose per frame pair; trajectory by chaining","training and evaluation on a single NVIDIA RTX 2080Ti with TensorFlow 1.9.0; inference time is not reported in the main paper (Sec. 4.2)","https:\u002F\u002Fgithub.com\u002FIRMVLab\u002FPWCLONet","MIT (LICENSE file read)",[48,52,56],{"relation":49,"title":50,"doi_or_url":51},"preprint","PWCLO-Net (arXiv v2, accepted CVPR 2021 version)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2012.00972",{"relation":53,"title":54,"doi_or_url":55},"follow_up_method","EfficientLO-Net (IEEE TPAMI; not read)","https:\u002F\u002Fdoi.org\u002F10.1109\u002FTPAMI.2022.3207015",{"relation":57,"title":58,"doi_or_url":45},"code_release","IRMVLab\u002FPWCLONet",{"id":5,"kind":60,"shortName":7,"title":8,"authors":61,"year":9,"venue":66,"venueType":67,"publisher":68,"volumeIssuePages":69,"doi":70,"arxivId":71,"url":72,"firstPublicDate":73,"publicationStatus":16,"metadataStatus":74,"fulltextStatus":15,"era":10,"classicReason":75,"codeUrl":45,"cluster":11,"topics":76,"mdpi":77,"verification":78,"label":6,"fulltextRoute":79,"versionRead":80,"addedByCensus":81},"method",[62,63,64,65],"Guangming Wang","Xinrui Wu","Zhe Liu","Hesheng Wang","2021 IEEE\u002FCVF Conference on Computer Vision and Pattern Recognition (CVPR)","conference","IEEE","pp. 15905-15914","10.1109\u002Fcvpr46437.2021.01565","2012.00972","https:\u002F\u002Fdoi.org\u002F10.1109\u002FCVPR46437.2021.01565","2020-12-02","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv v2 (2021-04-02), accepted CVPR 2021 version; IEEE version of record and supplementary material not read",true,[83,90],{"category":84,"model":85,"canonical":85,"role":86,"dataset":87,"specs":88,"locator":89},"lidar","Velodyne LiDAR (KITTI; model not specified in the paper)","dataset sensor","KITTI odometry","XYZ and reflectance provided; only XYZ used; transformed to the left camera frame","Sec. 4.1",{"category":91,"model":92,"canonical":93,"role":94,"dataset":95,"specs":96,"locator":97},"compute","NVIDIA RTX 2080Ti","NVIDIA RTX 2080 Ti","compute for runtime",null,"single GPU, TensorFlow 1.9.0, used for training and evaluation","Sec. 4.2",[],{"totalRows":100,"groupCount":101,"groups":102,"others":541},25,4,[103,424,465,503],{"slug":104,"group":105,"sourceId":5,"sourceLabel":6,"table":106,"selfRows":107,"metrics":108,"seqs":118,"entrants":144,"cells":171,"outcomes":418,"locators":419,"hardware":420,"wordings":421,"notes":422},"pwclonet2021-table-1","pwclonet2021:Table 1","Table 1",13,[109,113,115],{"label":110,"unit":111,"statistic":112,"alignment":75},"trel (average translational RMSE, %)","%","RMSE",{"label":114,"unit":111,"statistic":112,"alignment":75},"Mean on 07-10, trel",{"label":116,"unit":117,"statistic":112,"alignment":75},"Mean on 07-10, rrel (deg\u002F100m)","deg\u002F100m",[119,122,124,126,128,130,132,134,136,138,140,142],{"dataset":87,"sequence":120,"environment":121},"00* (training)","vehicle, road",{"dataset":87,"sequence":123,"environment":121},"01* (training)",{"dataset":87,"sequence":125,"environment":121},"02* (training)",{"dataset":87,"sequence":127,"environment":121},"03* (training)",{"dataset":87,"sequence":129,"environment":121},"04* (training)",{"dataset":87,"sequence":131,"environment":121},"05* (training)",{"dataset":87,"sequence":133,"environment":121},"06* (training)",{"dataset":87,"sequence":135,"environment":121},"07 (test)",{"dataset":87,"sequence":137,"environment":121},"08 (test)",{"dataset":87,"sequence":139,"environment":121},"09 (test)",{"dataset":87,"sequence":141,"environment":121},"10 (test)",{"dataset":87,"sequence":143,"environment":121},"mean on 07-10 (test)",[145,148,151,154,157,159,161,164,166,169],{"name":146,"methodId":147,"linkable":81,"proposed":77,"self":77},"Full LOAM [31]","loam2017_auro",{"name":149,"methodId":150,"linkable":81,"proposed":77,"self":77},"ICP-po2po","besl1992icp",{"name":152,"methodId":153,"linkable":81,"proposed":77,"self":77},"ICP-po2pl","chen1992pointtoplane",{"name":155,"methodId":156,"linkable":81,"proposed":77,"self":77},"GICP [19]","segal2009gicp",{"name":158,"methodId":95,"linkable":77,"proposed":77,"self":77},"CLS [21]",{"name":160,"methodId":95,"linkable":77,"proposed":77,"self":77},"Velas et al. [22]",{"name":162,"methodId":163,"linkable":81,"proposed":77,"self":77},"LO-Net [10]","lonet2019",{"name":165,"methodId":95,"linkable":77,"proposed":77,"self":77},"DMLO [11]",{"name":167,"methodId":168,"linkable":81,"proposed":77,"self":77},"LOAM w\u002Fo mapping (published code run by authors)","loam2014",{"name":170,"methodId":5,"linkable":81,"proposed":81,"self":81},"Ours (PWCLO-Net)",[172,176,179,182,185,187,190,193,196,199,202,205,208,210,212,214,216,218,220,222,224,226,228,230,232,234,236,238,240,241,243,245,247,249,251,253,255,257,259,261,263,265,267,269,271,273,275,277,279,281,283,285,287,289,291,292,294,296,298,299,301,303,304,306,308,310,312,314,316,318,319,321,323,324,326,327,329,331,333,335,337,339,341,342,344,346,348,350,352,354,356,358,360,361,363,365,367,369,371,373,375,377,379,380,382,384,386,388,390,392,394,396,398,400,402,404,406,408,410,412,414,416],[173,173,173,174,175,173,175,175,173],0,1.1,-1,[173,173,177,178,175,173,175,175,173],1,2.79,[173,173,180,181,175,173,175,175,173],2,1.54,[173,173,183,184,175,173,175,175,173],3,1.13,[173,173,101,186,175,173,175,175,173],1.45,[173,173,188,189,175,173,175,175,173],5,0.75,[173,173,191,192,175,173,175,175,173],6,0.72,[173,173,194,195,175,173,175,175,173],7,0.69,[173,173,197,198,175,173,175,175,173],8,1.18,[173,173,200,201,175,173,175,175,173],9,1.2,[173,173,203,204,175,173,175,175,173],10,1.51,[173,177,206,207,175,173,175,175,173],11,1.145,[173,180,206,209,175,173,175,175,173],0.498,[177,173,173,211,175,173,175,175,173],6.88,[177,173,177,213,175,173,175,175,173],11.21,[177,173,180,215,175,173,175,175,173],8.21,[177,173,183,217,175,173,175,175,173],11.07,[177,173,101,219,175,173,175,175,173],6.64,[177,173,188,221,175,173,175,175,173],3.97,[177,173,191,223,175,173,175,175,173],1.95,[177,173,194,225,175,173,175,175,173],5.17,[177,173,197,227,175,173,175,175,173],10.04,[177,173,200,229,175,173,175,175,173],6.93,[177,173,203,231,175,173,175,175,173],8.91,[177,177,206,233,175,173,175,175,173],7.763,[177,180,206,235,175,173,175,175,173],3.978,[180,173,173,237,175,173,175,175,173],3.8,[180,173,177,239,175,173,175,175,173],13.53,[180,173,180,200,175,173,175,175,173],[180,173,183,242,175,173,175,175,173],2.72,[180,173,101,244,175,173,175,175,173],2.96,[180,173,188,246,175,173,175,175,173],2.29,[180,173,191,248,175,173,175,175,173],1.77,[180,173,194,250,175,173,175,175,173],1.55,[180,173,197,252,175,173,175,175,173],4.42,[180,173,200,254,175,173,175,175,173],3.95,[180,173,203,256,175,173,175,175,173],6.13,[180,177,206,258,175,173,175,175,173],4.013,[180,180,206,260,175,173,175,175,173],1.968,[183,173,173,262,175,173,175,175,173],1.29,[183,173,177,264,175,173,175,175,173],4.39,[183,173,180,266,175,173,175,175,173],2.53,[183,173,183,268,175,173,175,175,173],1.68,[183,173,101,270,175,173,175,175,173],3.76,[183,173,188,272,175,173,175,175,173],1.02,[183,173,191,274,175,173,175,175,173],0.92,[183,173,194,276,175,173,175,175,173],0.64,[183,173,197,278,175,173,175,175,173],1.58,[183,173,200,280,175,173,175,175,173],1.97,[183,173,203,282,175,173,175,175,173],1.31,[183,177,206,284,175,173,175,175,173],1.375,[183,180,206,286,175,173,175,175,173],0.648,[101,173,173,288,175,173,175,175,173],2.11,[101,173,177,290,175,173,175,175,173],4.22,[101,173,180,246,175,173,175,175,173],[101,173,183,293,175,173,175,175,173],1.63,[101,173,101,295,175,173,175,175,173],1.59,[101,173,188,297,175,173,175,175,173],1.98,[101,173,191,274,175,173,175,175,173],[101,173,194,300,175,173,175,175,173],1.04,[101,173,197,302,175,173,175,175,173],2.14,[101,173,200,223,175,173,175,175,173],[101,173,203,305,175,173,175,175,173],3.46,[101,177,206,307,175,173,175,175,173],2.148,[101,180,206,309,175,173,175,175,173],0.995,[188,173,173,311,175,173,175,175,173],3.02,[188,173,177,313,175,173,175,175,173],4.44,[188,173,180,315,175,173,175,175,173],3.42,[188,173,183,317,175,173,175,175,173],4.94,[188,173,101,248,175,173,175,175,173],[188,173,188,320,175,173,175,175,173],2.35,[188,173,191,322,175,173,175,175,173],1.88,[188,173,194,248,175,173,175,175,173],[188,173,197,325,175,173,175,175,173],2.89,[188,173,200,317,175,173,175,175,173],[188,173,203,328,175,173,175,175,173],3.27,[188,177,206,330,175,173,175,175,173],3.218,[191,173,173,332,175,173,175,175,173],1.47,[191,173,177,334,175,173,175,175,173],1.36,[191,173,180,336,175,173,175,175,173],1.52,[191,173,183,338,175,173,175,175,173],1.03,[191,173,101,340,175,173,175,175,173],0.51,[191,173,188,300,175,173,175,175,173],[191,173,191,343,175,173,175,175,173],0.71,[191,173,194,345,175,173,175,175,173],1.7,[191,173,197,347,175,173,175,175,173],2.12,[191,173,200,349,175,173,175,175,173],1.37,[191,173,203,351,175,173,175,175,173],1.8,[191,177,206,353,175,173,175,175,173],1.748,[191,180,206,355,175,173,175,175,173],0.793,[194,173,194,357,175,173,175,175,173],0.73,[194,173,197,359,175,173,175,175,173],1.08,[194,173,200,174,175,173,175,175,173],[194,173,203,362,175,173,175,175,173],1.12,[194,177,206,364,175,173,175,175,173],1.008,[194,180,206,366,175,173,175,175,173],0.538,[197,173,173,368,175,173,175,175,173],15.99,[197,173,177,370,175,173,175,175,173],3.43,[197,173,180,372,175,173,175,175,173],9.4,[197,173,183,374,175,173,175,175,173],18.18,[197,173,101,376,175,173,175,175,173],9.59,[197,173,188,378,175,173,175,175,173],9.16,[197,173,191,231,175,173,175,175,173],[197,173,194,381,175,173,175,175,173],10.87,[197,173,197,383,175,173,175,175,173],12.72,[197,173,200,385,175,173,175,175,173],8.1,[197,173,203,387,175,173,175,175,173],12.67,[197,177,206,389,175,173,175,175,173],11.09,[197,180,206,391,175,173,175,175,173],6.405,[200,173,173,393,175,173,175,175,173],0.78,[200,173,177,395,175,173,175,175,173],0.67,[200,173,180,397,175,173,175,175,173],0.86,[200,173,183,399,175,173,175,175,173],0.76,[200,173,101,401,175,173,175,175,173],0.37,[200,173,188,403,175,173,175,175,173],0.45,[200,173,191,405,175,173,175,175,173],0.27,[200,173,194,407,175,173,175,175,173],0.6,[200,173,197,409,175,173,175,175,173],1.26,[200,173,200,411,175,173,175,175,173],0.79,[200,173,203,413,175,173,175,175,173],1.69,[200,177,206,415,175,173,175,175,173],1.085,[200,180,206,417,175,173,175,175,173],0.49,[],[106],[],[],[423],"KITTI odometry, trained on 00-06 (marked *) and tested on 07-10; trel = average translational RMSE (%) over 100-800 m subsequences; rows other than LOAM w\u002Fo mapping and Ours are copied from LO-Net [10]; LOAM is a full system with mapping, others are odometry only",{"slug":425,"group":426,"sourceId":5,"sourceLabel":6,"table":427,"selfRows":101,"metrics":428,"seqs":435,"entrants":440,"cells":444,"outcomes":459,"locators":460,"hardware":461,"wordings":462,"notes":463},"pwclonet2021-table-2","pwclonet2021:Table 2","Table 2",[429,431,433],{"label":430,"unit":111,"statistic":112,"alignment":75},"trel (%)",{"label":432,"unit":111,"statistic":112,"alignment":75},"Mean trel (%)",{"label":434,"unit":117,"statistic":112,"alignment":75},"Mean rrel (deg\u002F100m)",[436,437,438],{"dataset":87,"sequence":135,"environment":121},{"dataset":87,"sequence":137,"environment":121},{"dataset":87,"sequence":439,"environment":121},"mean on test sequences",[441,443],{"name":442,"methodId":95,"linkable":77,"proposed":77,"self":77},"LodoNet [32]",{"name":170,"methodId":5,"linkable":81,"proposed":81,"self":81},[445,447,449,450,452,454,455,457],[173,173,173,446,175,173,175,175,173],1.86,[173,173,177,448,175,173,175,175,173],2.04,[173,177,180,223,175,173,175,175,173],[173,180,180,451,175,173,175,175,173],1.305,[177,173,173,453,175,173,175,175,173],0.61,[177,173,177,262,175,173,175,175,173],[177,177,180,456,175,173,175,175,173],0.95,[177,180,180,458,175,173,175,175,173],0.5,[],[427],[],[],[464],"KITTI, trained on 00-06 and 09-10, tested on 07-08 to match LodoNet [32]",{"slug":466,"group":467,"sourceId":5,"sourceLabel":6,"table":468,"selfRows":101,"metrics":469,"seqs":473,"entrants":478,"cells":482,"outcomes":497,"locators":498,"hardware":499,"wordings":500,"notes":501},"pwclonet2021-table-3","pwclonet2021:Table 3","Table 3",[470,471,472],{"label":430,"unit":111,"statistic":112,"alignment":75},{"label":432,"unit":111,"statistic":112,"alignment":75},{"label":434,"unit":117,"statistic":112,"alignment":75},[474,476,477],{"dataset":87,"sequence":475,"environment":121},"04 (test)",{"dataset":87,"sequence":141,"environment":121},{"dataset":87,"sequence":439,"environment":121},[479,481],{"name":480,"methodId":95,"linkable":77,"proposed":77,"self":77},"DeepPCO [27]",{"name":170,"methodId":5,"linkable":81,"proposed":81,"self":81},[483,485,487,489,491,492,494,496],[173,173,173,484,175,173,175,175,173],2.63,[173,173,177,486,175,173,175,175,173],2.47,[173,177,180,488,175,173,175,175,173],2.55,[173,180,180,490,175,173,175,175,173],4.82,[177,173,173,357,175,173,175,175,173],[177,173,177,493,175,173,175,175,173],1.57,[177,177,180,495,175,173,175,175,173],1.15,[177,180,180,458,175,173,175,175,173],[],[468],[],[],[502],"KITTI, trained on 00-03 and 05-09, tested on 04 and 10 to match DeepPCO [27]",{"slug":504,"group":505,"sourceId":5,"sourceLabel":6,"table":506,"selfRows":101,"metrics":507,"seqs":511,"entrants":515,"cells":519,"outcomes":535,"locators":536,"hardware":537,"wordings":538,"notes":539},"pwclonet2021-table-4","pwclonet2021:Table 4","Table 4",[508,509,510],{"label":430,"unit":111,"statistic":112,"alignment":75},{"label":432,"unit":111,"statistic":112,"alignment":75},{"label":434,"unit":117,"statistic":112,"alignment":75},[512,513,514],{"dataset":87,"sequence":139,"environment":121},{"dataset":87,"sequence":141,"environment":121},{"dataset":87,"sequence":439,"environment":121},[516,518],{"name":517,"methodId":95,"linkable":77,"proposed":77,"self":77},"Cho et al. [3] (unsupervised)",{"name":170,"methodId":5,"linkable":81,"proposed":81,"self":81},[520,522,524,526,528,529,531,533],[173,173,173,521,175,173,175,175,173],4.87,[173,173,177,523,175,173,175,175,173],5.02,[173,177,180,525,175,173,175,175,173],4.945,[173,180,180,527,175,173,175,175,173],1.89,[177,173,173,357,175,173,175,175,173],[177,173,177,530,175,173,175,175,173],1.16,[177,177,180,532,175,173,175,175,173],0.945,[177,180,180,534,175,173,175,175,173],0.295,[],[506],[],[],[540],"KITTI, trained on 00-08, tested on 09 and 10 to match the unsupervised method of Cho et al. [3]",[],1790510663347]