DeiT III: Revenge of the ViT

Open paper
Benchmark Model Rank Results
semantic-segmentation-on-ade20k-valDeiT-L#28mIoU: 55.6
semantic-segmentation-on-ade20k-valDeiT-B#35mIoU: 54.1
image-classification-on-imagenet-realViT-L @384 (DeiT III, 21k)#59Top 1 Accuracy: 87.7%Number of params: 304M
image-classification-on-imagenet-realViT-H @224 (DeiT III, 21k)#60Top 1 Accuracy: 87.2%Number of params: 632M
image-classification-on-imagenet-realViT-L @224 (DeiT III, 21k)#61Top 1 Accuracy: 87.0%
image-classification-on-imagenetViT-B @384 (DeiT III, 21k)#136Top 1 Accuracy: 86.7%
image-classification-on-imagenetViT-L#206Top 1 Accuracy: 85.8%Number of params: 304.8MGFLOPs: 191.2
image-classification-on-imagenetViT-B @224 (DeiT III, 21k)#220Top 1 Accuracy: 85.7%
image-classification-on-imagenetViT-H @224 (DeiT III)#265Top 1 Accuracy: 85.2%
image-classification-on-imagenetViT-B @384 (DeiT III)#283Top 1 Accuracy: 85.0%Number of params: 87M
image-classification-on-imagenetViT-L @224 (DeiT III)#295Top 1 Accuracy: 84.9%
image-classification-on-imagenetViT-B @224 (DeiT III)#418Top 1 Accuracy: 83.8%
image-classification-on-imagenetViT-S @384 (DeiT III)#460Top 1 Accuracy: 83.4%Number of params: 22MGFLOPs: 15.5
image-classification-on-imagenetViT-S @224 (DeiT III, 21k)#497Top 1 Accuracy: 83.1%
image-classification-on-imagenetViT-S @224 (DeiT III)#681Top 1 Accuracy: 81.4%

Rank counts only results with a code link.