Point Cloud Segmentation on ScanNet v2 (val)
77.5mIoUPTv3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PTv3Architecture=Hybrid, PyTorch Native=✗, Params (M)=46.2, Hardware Platform=NVIDIA H100, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 77.5 | 0.562 | 83.73 | — | |
| PTv3Architecture=Hybrid, PyTorch Native=✗, Params (M)=46.2, Hardware Platform=Jetson Orin AGX, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 77.5 | 0.541 | 285.55 | — | |
| PTv3Architecture=Hybrid, PyTorch Native=✗, Params (M)=46.2, Hardware Platform=CPU (Intel Core i5-10400F), Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 77.5 | — | — | 3.48 | |
| PTXArchitecture=Transformer, PyTorch Native=✓, Params (M)=9.6, Hardware Platform=NVIDIA H100, Precision=BF16/FP8, FlashAttention-3=true, Test-time augmentation=true2026.04 | 76.5 | 0.253 | 52.52 | — | |
| PTXArchitecture=Transformer, PyTorch Native=✓, Params (M)=9.6, Hardware Platform=AMD MI300X, Precision=BF16, Test-time augmentation=true2026.04 | 76.5 | 0.39 | 53.98 | — | |
| PTXArchitecture=Transformer, PyTorch Native=✓, Params (M)=9.6, Hardware Platform=Jetson Orin AGX, Precision=BFloat16, Test-time augmentation=true2026.04 | 76.5 | 0.239 | 259.54 | — | |
| PTXArchitecture=Transformer, PyTorch Native=✓, Params (M)=9.6, Hardware Platform=CPU (Intel Core i5-10400F), Precision=BF16, Test-time augmentation=true2026.04 | 76.5 | — | — | 3.4 | |
| LitePT-SArchitecture=Hybrid, PyTorch Native=✗, Params (M)=12.7, Hardware Platform=NVIDIA H100, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.5 | 0.314 | 56.81 | — | |
| LitePT-SArchitecture=Hybrid, PyTorch Native=✗, Params (M)=12.7, Hardware Platform=Jetson Orin AGX, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.5 | 0.283 | 196.34 | — | |
| LitePT-SArchitecture=Hybrid, PyTorch Native=✗, Params (M)=12.7, Hardware Platform=CPU (Intel Core i5-10400F), Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.5 | — | — | 1.75 | |
| Swin3D-SArchitecture=Transformer, PyTorch Native=✗, Params (M)=28.2, Hardware Platform=NVIDIA H100, Manual update of CUDA kernels=true, Extra data=true, Test-time augmentation=true2026.04 | 76.4 | 0.525 | 81 | — | |
| Swin3D-SArchitecture=Transformer, PyTorch Native=✗, Params (M)=28.2, Hardware Platform=Jetson Orin AGX, Extra data=true, Test-time augmentation=true2026.04 | 76.4 | 0.803 | 2,007.28 | — | |
| OACNNArchitecture=Convolution, PyTorch Native=✗, Params (M)=51.5, Hardware Platform=NVIDIA H100, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.1 | 0.708 | 132.89 | — | |
| OACNNArchitecture=Convolution, PyTorch Native=✗, Params (M)=51.5, Hardware Platform=Jetson Orin AGX, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.1 | 0.566 | 489.27 | — | |
| OACNNArchitecture=Convolution, PyTorch Native=✗, Params (M)=51.5, Hardware Platform=CPU (Intel Core i5-10400F), Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 76.1 | — | — | 1.65 | |
| OctformerArchitecture=Hybrid, PyTorch Native=✓, Params (M)=44.0, Hardware Platform=NVIDIA H100, CUDA kernels with PyTorch fallback=true, Test-time augmentation=true2026.04 | 75.7 | 2.073 | 58.23 | — | |
| OctformerArchitecture=Hybrid, PyTorch Native=✓, Params (M)=44.0, Hardware Platform=AMD MI300X, CUDA kernels with PyTorch fallback=true, Test-time augmentation=true2026.04 | 75.7 | 1.585 | 72.33 | — | |
| OctformerArchitecture=Hybrid, PyTorch Native=✓, Params (M)=44.0, Hardware Platform=Jetson Orin AGX, CUDA kernels with PyTorch fallback=true, Test-time augmentation=true2026.04 | 75.7 | 2.038 | 426.17 | — | |
| OctformerArchitecture=Hybrid, PyTorch Native=✓, Params (M)=44.0, Hardware Platform=CPU (Intel Core i5-10400F), CUDA kernels with PyTorch fallback=true, Test-time augmentation=true2026.04 | 75.7 | — | — | 6.68 | |
| PTv2Architecture=Transformer, PyTorch Native=✗, Params (M)=11.3, Hardware Platform=NVIDIA H100, Manual update of CUDA kernels=true, Test-time augmentation=true2026.04 | 75.4 | 1.652 | 173.16 | — | |
| PTv2Architecture=Transformer, PyTorch Native=✗, Params (M)=11.3, Hardware Platform=Jetson Orin AGX, Test-time augmentation=true2026.04 | 75.4 | 1.625 | 948.41 | — | |
| STArchitecture=Transformer, PyTorch Native=✗, Params (M)=18.8, Test-time augmentation=true2026.04 | 74.3 | — | — | — | |
| MinkUNet-34CArchitecture=Convolution, PyTorch Native=✗, Params (M)=37.9, Hardware Platform=NVIDIA H100, Manual update of CUDA kernels=true, Test-time augmentation=true2026.04 | 72.2 | 0.525 | 81 | — | |
| MinkUNet-34CArchitecture=Convolution, PyTorch Native=✗, Params (M)=37.9, Hardware Platform=Jetson Orin AGX, Test-time augmentation=true2026.04 | 72.2 | 0.503 | 465.84 | — | |
| MinkUNet-34CArchitecture=Convolution, PyTorch Native=✗, Params (M)=37.9, Hardware Platform=CPU (Intel Core i5-10400F), Test-time augmentation=true2026.04 | 72.2 | — | — | 5.68 | |
| PTv1Architecture=Transformer, PyTorch Native=✗, Params (M)=7.8, Hardware Platform=NVIDIA H100, Manual update of CUDA kernels=true, Test-time augmentation=true2026.04 | 70.6 | 0.749 | 1,036.21 | — | |
| PTv1Architecture=Transformer, PyTorch Native=✗, Params (M)=7.8, Hardware Platform=Jetson Orin AGX, Test-time augmentation=true2026.04 | 70.6 | 0.722 | 2,288.57 | — | |
| SpUNetArchitecture=Convolution, PyTorch Native=✗, Params (M)=39.2, Hardware Platform=NVIDIA H100, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 69.3 | 0.487 | 43.02 | — | |
| SpUNetArchitecture=Convolution, PyTorch Native=✗, Params (M)=39.2, Hardware Platform=Jetson Orin AGX, Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 69.3 | 0.464 | 138.58 | — | |
| SpUNetArchitecture=Convolution, PyTorch Native=✗, Params (M)=39.2, Hardware Platform=CPU (Intel Core i5-10400F), Implementation adapted for CPU=true, Test-time augmentation=true2026.04 | 69.3 | — | — | 1.8 |