4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks

Benchmark Model Rank Results
3d-semantic-segmentation-on-scannet-1SpUNet (MinkowskiNet)#7Top-1 IoU: 0.456Top-3 IoU: 0.683
3d-semantic-segmentation-on-scannet200MinkUNet#15val mIoU: 25.0test mIoU: 25.3
3d-semantic-segmentation-on-scribblekittiMinkowskiNet#3mIoU: 55.0
3d-semantic-segmentation-on-stpls3dMinkowskiNet#3mIOU: 51.3
3d-semantic-segmentation-on-wildscenesMinkUNet#3mIoU: 36.53mIoU (Temporal DA): 27.20mIoU (Env DA): 30.78
robust-3d-semantic-segmentation-on-nuscenes-cMinkUNet-34#2mean Corruption Error (mCE): 96.37%
robust-3d-semantic-segmentation-on-nuscenes-cMinkUNet-18#5mean Corruption Error (mCE): 100.00%
robust-3d-semantic-segmentation-on-robo3dMinkUNet-18#3mean Corruption Error (mCE): 100.00%
robust-3d-semantic-segmentation-on-robo3dMinkUNet-34#5mean Corruption Error (mCE): 100.61%
robust-3d-semantic-segmentation-on-wod-cMinkUNet-34#1mean Corruption Error (mCE): 96.21%
robust-3d-semantic-segmentation-on-wod-cMinkUNet-18#3mean Corruption Error (mCE): 100.00%
semantic-segmentation-on-s3disMinkowskiNet#35Mean IoU: 65.4Number of params: 37.9MParams (M): 37.9
semantic-segmentation-on-s3dis-area5MinkowskiNet#34mIoU: 65.4mAcc: 71.7Number of params: 37.9M
semantic-segmentation-on-scannetMinkowskiNet#25val mIoU: 72.2test mIoU: 73.4