Efficient Multi-Task Scene Analysis with RGB-D Transformers

Benchmark Model Rank Results
semantic-segmentation-on-scannetv2EMSAFormer#5Mean IoU: 56.4%
semantic-segmentation-on-sun-rgbdEMSANet (2x ResNet-34 NBt1D, PanopticNDT version, finetuned)#28Mean IoU: 48.82%
semantic-segmentation-on-nyu-depth-v2EMSAFormer (SwinV2-T-128-Multi-Aug)#41Mean IoU: 51.26%