Swin3D: A Pretrained Transformer Backbone for 3D Indoor Scene Understanding

Benchmark Model Rank Results
3d-object-detection-on-s3disSwin3D-L+FCAF3D#4mAP@0.5: 54.0mAP@0.25: 72.1
semantic-segmentation-on-s3disSwin3D-L#4Mean IoU: 79.8mAcc: 88.0oAcc: 92.4Number of params: N/A
semantic-segmentation-on-s3dis-area5Swin3D-L#4mIoU: 74.5mAcc: 80.5oAcc: 92.7Number of params: N/A
3d-object-detection-on-scannetv2Swin3D-L+CAGroup3D#6mAP@0.5: 63.2mAP@0.25: 76.4
semantic-segmentation-on-scannetSwin3D-L#7val mIoU: 77.5test mIoU: 77.9