MultiMAE: Multi-modal Multi-task Masked Autoencoders

Benchmark Model Rank Results
semantic-segmentation-on-hypersimMultiMAE (ViT-B)#2mIoU: 37.0
semantic-segmentation-on-hypersimMAE (ViT-B)#3mIoU: 36.5
semantic-segmentation-on-hypersimDINO (ViT-B)#4mIoU: 32.5
semantic-segmentation-on-hypersimMoCo-v3 (ViT-B)#5mIoU: 31.7
semantic-segmentation-on-nyu-depth-v2MultiMAE (ViT-B)#19Mean IoU: 56.0%
semantic-segmentation-on-ade20k-valMultiMAE (ViT-B)#68mIoU: 46.2
semantic-segmentation-on-ade20kMultiMAE (ViT-B)#195Validation mIoU: 46.2