3D Object Detection on ScanNet V2 (Accuracy Metrics)
51.9AP50GroupFree3D + Bridge3D
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GroupFree3D + Bridge3DPre-trained=true2023.05 | 51.9 | 69.1 | |
| GroupFree3D + PIMAEPre-trained=true2023.05 | 50.5 | 67.9 | |
| GroupFree3D + Point-MAEPre-trained=true2023.05 | 49.8 | 67.4 | |
| 3D-MPAArch.=VoteNet + GraphConv2021.09 | 49.2 | 64.2 | |
| MM3DScene + H3DNetModel Size=6.87M, Train Time=36.1h2022.12 | 48.9 | 66.8 | |
| GroupFree3DPre-trained=false2023.05 | 48.9 | 67.3 | |
| Group-Free (L6, O256)Backbone=PN++w2 ×2026.04 | 48.9 | 67.3 | |
| Group-Free (L6, O256)Backbone=PN++w2 ×2026.04 | 48.9 | 67.3 | |
| GroupFree3D + Bridge3D (from scratch)Pre-trained=false2023.05 | 48.3 | 66.1 | |
| H3DNetArch.=VoteNet + 3D primitives2021.09 | 48.1 | 67.2 | |
| H3DNet (scratch)Model Size=4.74M, Train Time=12.3h, Infer Time=7.3s2022.12 | 47.4 | 64.8 | |
| 3DETR-mArch.=Tx.2021.09 | 47 | 65 | |
| 3DETR-m (paper)Backbone=SA12026.04 | 47 | 65 | |
| 3DETR-m (our baseline)Backbone=SA12026.04 | 46.58 | 64.88 | |
| 3DETR-m + MSABackbone=SA12026.04 | 45.35 | 65.67 | |
| 3DETR + Bridge3DPre-trained=true2023.05 | 44.2 | 65.3 | |
| 3DETR + MSABackbone=SA12026.04 | 43.51 | 62.32 | |
| DepthContrastPre-trained=true2023.05 | 42.9 | 64 | |
| ACTSSL=true, Input=xyz2022.12 | 42.1 | 63.8 | |
| MM3DScene + VoteNetModel Size=1.48M, Train Time=15.6h2022.12 | 41.5 | 63.1 | |
| MM-3DScene + VoteNetBackbone=VoteNet, Pre-training approach=MM-3DScene2022.12 | 41.5 | 63.1 | |
| DPCoPre-trained=true2023.05 | 41.5 | 64.2 | |
| MLCVNetBackbone=PN++2026.04 | 41.4 | 64.5 | |
| MLCVNetBackbone=PN++2026.04 | 41.4 | 64.5 | |
| MM-3DScene (w/o LCSD) + VoteNetBackbone=VoteNet, Pre-training approach=MM-3DScene (w/o LCSD)2022.12 | 41.3 | 61.9 | |
| 3DETR + ACTPre-trained=true2023.05 | 41 | 63.5 | |
| MaskPointSSL=true, Input=xyz2022.12 | 40.6 | 63.4 | |
| 3DETR + Point-MAEPre-trained=true2023.05 | 40.6 | 63.4 | |
| 3DETR + MaskPointPre-trained=true2023.05 | 40.6 | 63.4 | |
| 3DETR + PIMAEPre-trained=true2023.05 | 40.2 | 63 | |
| 4DContrast + VoteNetBackbone=VoteNet, Pre-training approach=4DContrast2022.12 | 40 | — | |
| 4DContrastPre-trained=true2023.05 | 40 | — | |
| CSC + VoteNetBackbone=VoteNet, Pre-training approach=CSC2022.12 | 39.3 | — | |
| Hou et al.Pre-trained=true2023.05 | 39.3 | — | |
| 3DETR (our baseline)Backbone=SA12026.04 | 38.73 | 61.33 | |
| 3DETR + Bridge3D (from scratch)Pre-trained=false2023.05 | 38.6 | 61.1 | |
| STRLSSL=true, Input=xyz2022.12 | 38.4 | 59.5 | |
| Point-BERTSSL=true, Input=xyz2022.12 | 38.3 | 61 | |
| 3DETR + Point-BERTPre-trained=true2023.05 | 38.3 | 61 | |
| PointContrastSSL=true, Input=xyz2022.12 | 38 | 59.2 | |
| PointContrast + VoteNetBackbone=VoteNet, Pre-training approach=PointContrast2022.12 | 38 | 58.5 | |
| PointContrastPre-trained=true2023.05 | 38 | 59.2 | |
| 3DETRSSL=false, Input=xyz2022.12 | 37.9 | 62.1 | |
| 3DETRPre-trained=false2023.05 | 37.9 | 62.1 | |
| 3DETRArch.=Tx.2021.09 | 37.5 | 62.7 | |
| VoteNetArch.=VoteNet2021.09 | 37.5 | 60.4 | |
| 3DETR (paper)Backbone=SA12026.04 | 37.5 | 62.7 | |
| RandomRoomsSSL=true, Input=xyz2022.12 | 36.2 | 61.3 | |
| RandomRooms + VoteNetBackbone=VoteNet, Pre-training approach=RandomRooms2022.12 | 36.2 | 61.3 | |
| VoteNet (scratch)Model Size=0.95M, Train Time=5.9h, Infer Time=0.2s2022.12 | 35.4 | 58.7 | |
| VoteNetPre-training approach=scratch2022.12 | 35.1 | 58.7 | |
| HGNetArch.=VoteNet + GraphConv2021.09 | 34.4 | 61.3 | |
| VoteNetSSL=false, Input=xyz2022.12 | 33.5 | 58.6 | |
| VoteNetInput=PC2023.08 | 33.5 | 58.6 | |
| VoteNetPre-trained=false2023.05 | 33.5 | 58.6 | |
| VoteNetBackbone=PN++2026.04 | 33.5 | 58.6 | |
| VoteNetBackbone=PN++2026.04 | 33.5 | 58.6 | |
| ImGeoNetInput=RGB2023.08 | 28.4 | 54.8 | |
| ImVoxelNetInput=RGB2023.08 | 23.8 | 48.7 | |
| 3D-SIS2022.12 | 22.5 | 40.2 | |
| 3D-SISInput=RGB + PC2023.08 | 22.5 | 40.2 | |
| BoxNetArch.=BoxNet2021.09 | 21.1 | 49 | |
| GSPN2022.12 | 17.7 | 30.6 | |
| 3D-SISInput=PC2023.08 | 14.6 | 25.4 | |
| F-PointNet2022.12 | 10.8 | 19.8 | |
| DSS2022.12 | 6.8 | 15.2 | |
| DepthContrastSSL=true, Input=xyz2022.12 | — | 61.3 |