Link Search Menu Expand Document

3D Classification

Track 1: 3D classification (No Feat)

In this track, the models are trained only using voxel coordinates on PeRFception-CO3D. In other words, using spherical harmonic coefficients and density values are prohibited. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.

ModelAcc@1Acc@5CheckpointCode
Mink-ResNet10166.21 \(\pm\) 0.9386.50 \(\pm\) 0.09linklink
Mink-ResNet5065.25 \(\pm\) 0.7585.71 \(\pm\) 0.64linklink
Mink-ResNet3464.55 \(\pm\) 0.8484.26 \(\pm\) 0.47linklink
Mink-ResNet1863.85 \(\pm\) 0.3384.47 \(\pm\) 0.54linklink
Mink-ResNet1459.36 \(\pm\) 0.3081.30 \(\pm\) 0.87linklink

Track 2: 3D classification (SH)

In this track, the models are trained with spherical harmonic coefficients and voxel coordinates on PeRFception-CO3D. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.

ModelAcc@1Acc@5CheckpointCode
Mink-ResNet10178.04 \(\pm\) 0.5889.60 \(\pm\) 0.36linklink
Mink-ResNet5077.59 \(\pm\) 0.1790.98 \(\pm\) 0.40linklink
Mink-ResNet3476.51 \(\pm\) 0.6191.79 \(\pm\) 0.49linklink
Mink-ResNet1875.58 \(\pm\) 0.3792.73 \(\pm\) 0.21linklink
Mink-ResNet1471.87 \(\pm\) 0.6193.24 \(\pm\) 0.49linklink

Track 3: 3D classification (D)

In this track, the models are trained with density values and voxel coordinates on PeRFception-CO3D. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.

ModelAcc@1Acc@5CheckpointCode
Mink-ResNet10177.27 \(\pm\) 0.6192.68 \(\pm\) 0.34linklink
Mink-ResNet5076.42 \(\pm\) 0.1992.91 \(\pm\) 0.24linklink
Mink-ResNet3476.38 \(\pm\) 0.3491.79 \(\pm\) 0.61linklink
Mink-ResNet1875.18 \(\pm\) 0.7091.10 \(\pm\) 0.24linklink
Mink-ResNet1472.44 \(\pm\) 0.2990.04 \(\pm\) 0.13linklink

Track 4: 3D classification (SH + D)

In this track, spherical harmonic coefficients, density, and voxel coordinates are available. We sort the models based on the Acc@1 score. Models are trained for three times. We report mean and std for each model.

ModelAcc@1Acc@5CheckpointCode
Mink-ResNet5077.53 \(\pm\) 0.2792.93 \(\pm\) 0.54linklink
Mink-ResNet10177.19 \(\pm\) 0.8993.09 \(\pm\) 0.21linklink
Mink-ResNet3476.50 \(\pm\) 0.0391.98 \(\pm\) 0.08linklink
Mink-ResNet1875.72 \(\pm\) 0.2591.54 \(\pm\) 0.21linklink
Mink-ResNet1472.92 \(\pm\) 0.4290.83 \(\pm\) 0.03linklink