List your company
ResearchFormerly WizWand

Zero-shot Classification on ModelNet40

85.4AccuracyE-3DSNN-H
1.88823.56945.2566.931Nov 21, 2022Apr 22, 2023Sep 21, 2023Feb 21, 2024Jul 22, 2024Dec 21, 2024May 23, 2025
Updated 5mo ago

Evaluation results

MethodLinks
E-3DSNN-HArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Voxel, T × D=1×4, Point+Text Param (M)=46.7, Energy (mJ)=0.792025.0585.4
Point-BertArchitecture=ANN, Pre-train Method=Ulip2 (Xue et al., 2024), Input=Point, T × D=N/A, Point+Text Param (M)=21.9+202.5, Energy (mJ)=83.7+71.72025.0584.7
E-3DSNN-LArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Voxel, T × D=1×4, Point+Text Param (M)=17.7, Energy (mJ)=0.642025.0584.6
SparseconvArchitecture=ANN, Pre-train Method=Openshape (Liu et al., 2023b), Input=Voxel, T × D=N/A, Point+Text Param (M)=41.3+202.5, Energy (mJ)=2.13+71.72025.0583.4
Spike-driven PointFormer-LArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Point, T × D=1×4, Point+Text Param (M)=22.1, Energy (mJ)=9.42025.0583.1
Point-BertArchitecture=ANN, Pre-train Method=Openshape (Liu et al., 2023b), Input=Point, T × D=N/A, Point+Text Param (M)=21.9+202.5, Energy (mJ)=83.2+71.72025.0582.8
Spike-driven PointFormer-SArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Point, T × D=1×4, Point+Text Param (M)=7.69, Energy (mJ)=5.12025.0582.1
E-3DSNN-SArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Voxel, T × D=1×4, Point+Text Param (M)=3.51, Energy (mJ)=0.092025.0581.3
E-3DSNN-TArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Voxel, T × D=1×4, Point+Text Param (M)=2.10, Energy (mJ)=0.042025.0579.6
SparseconvArchitecture=ANN, Pre-train Method=Openshape (Liu et al., 2023b), Input=Voxel, T × D=N/A, Point+Text Param (M)=5.3+202.5, Energy (mJ)=0.61+71.72025.0578.8
Spike PointNetArchitecture=SNN, Pre-train Method=SVL (Ours), Input=Point, T × D=1×4, Point+Text Param (M)=3.57, Energy (mJ)=0.272025.0576.3
PointNetArchitecture=ANN, Pre-train Method=Openshape (Liu et al., 2023b), Input=Point, T × D=N/A, Point+Text Param (M)=3.47+202.5, Energy (mJ)=20.1+71.72025.0574.9
Point-BertArchitecture=ANN, Pre-train Method=Ulip (Xue et al., 2023), Input=Point, T × D=N/A, Point+Text Param (M)=21.9+227.8, Energy (mJ)=81.2+80.62025.0569.6
PointCLIP V2Backbone=ViT-B/32, Hardware=1 RTX A6000, GFLOPS=16.51, Memory (GB)=29672022.1155.92
CLIP2PointBackbone=ViT-B/32, Hardware=1 RTX A6000, GFLOPS=16.46, Memory (GB)=30062022.1149.38
PointCLIPArchitecture=ANN, Pre-train Method=N/A, Input=Image, T × D=N/A, Point+Text Param (M)=25.5+57.3, Energy (mJ)=24.9+20.32025.0520.2
PointCLIPBackbone=ViT-B/32, Hardware=1 RTX A6000, GFLOPS=16.46, Memory (GB)=29012022.1116.94
SpikeCLIP*Architecture=SNN, Pre-train Method=N/A, Input=Image, T × D=4×1, Point+Text Param (M)=9.5+22.8, Energy (mJ)=10.6+0.412025.055.1