List your company
ResearchFormerly WizWand

Semantic Segmentation on ScanNet v2 (test)

79.4mIoUPTv3
71.0873.2475.477.56Mar 28, 2022Sep 28, 2022Apr 1, 2023Oct 3, 2023Apr 4, 2024Oct 6, 2024Apr 9, 2025
Updated 8mo ago

Evaluation results

MethodLinks
PTv3Training Strategy=multi-dataset joint training (PPT)2023.1279.4
PTv3 + PPTYear=2024, Training Regime=pre-trained2023.1279.4
PTv3 + PPTYear=2024, Training Status=Pre-trained, Backbone=PTv32023.0879.4
PPT (sup.)Year=2023, variant=enhanced2025.0379.4
Masked Scene ModelingBackbone=HUNet, Pre-training=Masked Scene Modeling, Fine-tuning=true2025.0478.5
MSCBackbone=HUNet, Pre-training=MSC, Fine-tuning=true2025.0478.2
Mix3DEnsemble voting=true2023.0478.1
SWIN3Dn-LPre-training strategy=pre-trained, Point Normals=true, Ensemble voting=true, Backbone scale=Large2023.0477.9
PTv3Training Strategy=trained from scratch2023.1277.9
Swin3DYear=2023, Training Regime=scratch2023.1277.9
Swin3D + SupervisedYear=2023, Training Regime=pre-trained2023.1277.9
PTv3Year=2024, Training Regime=scratch2023.1277.9
Swin3D + SupervisedYear=2023, Training Status=Pre-trained, Backbone=Swin3D2023.0877.9
PTv3Year=2024, Training Status=Scratch2023.0877.9
Swin3DYear=20232025.0377.9
SupervisedYear=2023, parent_method=Swin3D2025.0377.9
PTv3Year=20232025.0377.9
ODIN-Swin-BInput Source=Sensor RGBD Point Cloud, Backbone=Swin-B2024.0177.8
HUNetBackbone=HUNet, Pre-training=Scratch, Fine-tuning=true2025.0477
Swin3D-LInput Source=Mesh Sampled Point Cloud2024.0176.7
OctFormerEnsemble voting=true2023.0476.6
PointHR#Params=7.1M, FLOPs=10.3G2023.1076.6
PointHR-LMemory usage (GB)=23.582023.1076.6
OctFormerTraining Strategy=trained from scratch2023.1276.6
OctFormerYear=2023, Training Regime=scratch2023.1276.6
MinkUNet + PPTYear=2024, Training Regime=pre-trained2023.1276.6
OctFormerYear=2023, Training Status=Scratch2023.0876.6
MinkUNet + PPTYear=2024, Training Status=Pre-trained, Backbone=MinkUNet2023.0876.6
OctFormerYear=20232025.0376.6
PPT (sup.)Year=2023, variant=baseline2025.0376.6
VMVFInput Source=Rendered RGBD Point Cloud2024.0176.4
O-CNNEnsemble voting=true2023.0476.2
OctFormerInput Source=Mesh Sampled Point Cloud2024.0175.7
GCBackbone=SR-UNet, Pre-training=GC, Fine-tuning=true2025.0475.7
OA-CNNsYear=2024, Training Regime=scratch2023.1275.6
OA-CNNsYear=2024, Training Status=Scratch2023.0875.6
ConDaFormerInput=point, Test-time augmentation (TTA)=true2023.1275.5
Point Transformer v2Input Source=Mesh Sampled Point Cloud2024.0175.4
MSCBackbone=SR-UNet, Pre-training=MSC, Fine-tuning=true2025.0475.3
PTv2Input=point2022.1075.2
PointTransformerV2# Params(M)=12.8, Input=point, Runtime(ms)=2662022.0875.2
PointTransformerV2Ensemble voting=true2023.0475.2
PTv2#Params=11.3M, FLOPs=14.3G2023.1075.2
PTv2Memory usage (GB)=21.502023.1075.2
PTv2Input=point, Test-time augmentation (TTA)=true2023.1275.2
PointConvFormer# Params(M)=9.4, Input=Point, Runtime(ms)=145.52022.0874.9
PointConvFormerEnsemble voting=true2023.0474.9
BPNet2023.1074.9
PointConvFormerYear=2023, Training Regime=scratch2023.1274.9
PointConvFormerYear=2023, Training Status=Scratch2023.0874.9
PointConvFormerYear=20232025.0374.9
Stratified Transformer# Params(M)=18.8, Input=point, Runtime(ms)=1689.32022.0874.7
Stratified TransformerEnsemble voting=true2023.0474.7
StraFormer#Params=8.02M, FLOPs=12.4G2023.1074.7
ConDaFormerInput=point, Test-time augmentation (TTA)=false2023.1274.7
ODINYear=20242025.0374.4
Stratified TransformerInput Source=Mesh Sampled Point Cloud2024.0174.3
SAT2023.0174.2
PTv2Training Strategy=trained from scratch2023.1274.2
PTv2Year=2022, Training Regime=scratch2023.1274.2
PTv2Year=2022, Training Status=Scratch2023.0874.2
PTv2Year=20222025.0374.2
PCBackbone=SR-UNet, Pre-training=PC, Fine-tuning=true2025.0474.1
LargeKernel3DEnsemble voting=true2023.0473.9
LargeKernelInput=voxel, Test-time augmentation (TTA)=false2023.1273.9
LargeKernel3DYear=2023, Training Regime=scratch2023.1273.9
LargeKernel3DYear=2023, Training Status=Scratch2023.0873.9
LargeKernel3DYear=20232025.0373.9
CSCBackbone=SR-UNet, Pre-training=CSC, Fine-tuning=true2025.0473.8
Stratified TransformerInput=point2022.0373.7
Stratified Transformer2023.0173.7
StratifiedInput=point, Test-time augmentation (TTA)=false2023.1273.7
STTraining Strategy=trained from scratch2023.1273.7
Stratified TransformerYear=2022, Training Regime=scratch2023.1273.7
Stratified TransformerYear=2022, Training Status=Scratch2023.0873.7
SratifiedTranformerYear=20222025.0373.7
MinkowskiNetInput=voxel2022.0373.6
MinkUNetInput=voxel2022.1073.6
MinkowskiNet2023.0173.6
MinkowskiNet42# Params(M)=37.9, Input=Voxel, Runtime(ms)=115.62022.0873.6
MinkUNet2023.1073.6
MinkUNetInput=voxel, Test-time augmentation (TTA)=false2023.1273.6
MinkUNetTraining Strategy=multi-dataset joint training (PPT)2023.1273.6
MinkUNetYear=2019, Training Regime=scratch2023.1273.6
MinkUNetYear=2019, Training Status=Scratch2023.0873.6
SparseUNetYear=20192025.0373.6
ODIN-ResNet50Input Source=Sensor RGBD Point Cloud, Backbone=ResNet502024.0173.3
SparseConvNetInput=voxel2022.0372.5
SparseConvNetInput=voxel2022.1072.5
SparseConvNet2023.0172.5
SparseConvNetInput=Voxel2022.0872.5
SparseCNN2023.1072.5
SparseConvNetInput=voxel, Test-time augmentation (TTA)=false2023.1272.5
SparseConvNetYear=2018, Training Regime=scratch2023.1272.5
SparseConvNetYear=2018, Training Status=Scratch2023.0872.5
SparseConvNetYear=20182025.0372.5
SR-UNetBackbone=SR-UNet, Pre-training=Scratch, Fine-tuning=true2025.0472.2
PointMetaBase-XXLEnsemble voting=true2023.0471.4
PMeta#Params=19.7M, FLOPs=11.0G2023.1071.4
PMetaMemory usage (GB)=20.382023.1071.4