List your company

3D Object Detection on ScanNet V2 (Accuracy Metrics)

51.9AP50GroupFree3D + Bridge3D
4.99617.17329.3541.527Sep 16, 2021Jun 22, 2022Mar 28, 2023Jan 1, 2024Oct 6, 2024Jul 12, 2025Apr 17, 2026

Evaluation results

MethodLinks
GroupFree3D + Bridge3DPre-trained=true2023.0551.969.1
GroupFree3D + PIMAEPre-trained=true2023.0550.567.9
GroupFree3D + Point-MAEPre-trained=true2023.0549.867.4
3D-MPAArch.=VoteNet + GraphConv2021.0949.264.2
MM3DScene + H3DNetModel Size=6.87M, Train Time=36.1h2022.1248.966.8
GroupFree3DPre-trained=false2023.0548.967.3
Group-Free (L6, O256)Backbone=PN++w2 ×2026.0448.967.3
Group-Free (L6, O256)Backbone=PN++w2 ×2026.0448.967.3
GroupFree3D + Bridge3D (from scratch)Pre-trained=false2023.0548.366.1
H3DNetArch.=VoteNet + 3D primitives2021.0948.167.2
H3DNet (scratch)Model Size=4.74M, Train Time=12.3h, Infer Time=7.3s2022.1247.464.8
3DETR-mArch.=Tx.2021.094765
3DETR-m (paper)Backbone=SA12026.044765
3DETR-m (our baseline)Backbone=SA12026.0446.5864.88
3DETR-m + MSABackbone=SA12026.0445.3565.67
3DETR + Bridge3DPre-trained=true2023.0544.265.3
3DETR + MSABackbone=SA12026.0443.5162.32
DepthContrastPre-trained=true2023.0542.964
ACTSSL=true, Input=xyz2022.1242.163.8
MM3DScene + VoteNetModel Size=1.48M, Train Time=15.6h2022.1241.563.1
MM-3DScene + VoteNetBackbone=VoteNet, Pre-training approach=MM-3DScene2022.1241.563.1
DPCoPre-trained=true2023.0541.564.2
MLCVNetBackbone=PN++2026.0441.464.5
MLCVNetBackbone=PN++2026.0441.464.5
MM-3DScene (w/o LCSD) + VoteNetBackbone=VoteNet, Pre-training approach=MM-3DScene (w/o LCSD)2022.1241.361.9
3DETR + ACTPre-trained=true2023.054163.5
MaskPointSSL=true, Input=xyz2022.1240.663.4
3DETR + Point-MAEPre-trained=true2023.0540.663.4
3DETR + MaskPointPre-trained=true2023.0540.663.4
3DETR + PIMAEPre-trained=true2023.0540.263
4DContrast + VoteNetBackbone=VoteNet, Pre-training approach=4DContrast2022.1240—
4DContrastPre-trained=true2023.0540—
CSC + VoteNetBackbone=VoteNet, Pre-training approach=CSC2022.1239.3—
Hou et al.Pre-trained=true2023.0539.3—
3DETR (our baseline)Backbone=SA12026.0438.7361.33
3DETR + Bridge3D (from scratch)Pre-trained=false2023.0538.661.1
STRLSSL=true, Input=xyz2022.1238.459.5
Point-BERTSSL=true, Input=xyz2022.1238.361
3DETR + Point-BERTPre-trained=true2023.0538.361
PointContrastSSL=true, Input=xyz2022.123859.2
PointContrast + VoteNetBackbone=VoteNet, Pre-training approach=PointContrast2022.123858.5
PointContrastPre-trained=true2023.053859.2
3DETRSSL=false, Input=xyz2022.1237.962.1
3DETRPre-trained=false2023.0537.962.1
3DETRArch.=Tx.2021.0937.562.7
VoteNetArch.=VoteNet2021.0937.560.4
3DETR (paper)Backbone=SA12026.0437.562.7
RandomRoomsSSL=true, Input=xyz2022.1236.261.3
RandomRooms + VoteNetBackbone=VoteNet, Pre-training approach=RandomRooms2022.1236.261.3
VoteNet (scratch)Model Size=0.95M, Train Time=5.9h, Infer Time=0.2s2022.1235.458.7
VoteNetPre-training approach=scratch2022.1235.158.7
HGNetArch.=VoteNet + GraphConv2021.0934.461.3
VoteNetSSL=false, Input=xyz2022.1233.558.6
VoteNetInput=PC2023.0833.558.6
VoteNetPre-trained=false2023.0533.558.6
VoteNetBackbone=PN++2026.0433.558.6
VoteNetBackbone=PN++2026.0433.558.6
ImGeoNetInput=RGB2023.0828.454.8
ImVoxelNetInput=RGB2023.0823.848.7
3D-SIS2022.1222.540.2
3D-SISInput=RGB + PC2023.0822.540.2
BoxNetArch.=BoxNet2021.0921.149
GSPN2022.1217.730.6
3D-SISInput=PC2023.0814.625.4
F-PointNet2022.1210.819.8
DSS2022.126.815.2
DepthContrastSSL=true, Input=xyz2022.12—61.3