ViTAEv2: Vision Transformer Advanced by Exploring Inductive Bias for Image Recognition and Beyond

Open paper
Benchmark Model Rank Results
image-classification-on-imagenet-realViTAE-H (MAE, 512)#2Accuracy: 91.2%Params: 644M
image-classification-on-cifar-100ViTAE-S#34Percentage correct: 90.8
image-classification-on-cifar-10ViTAE-S#40Percentage correct: 98.8
image-classification-on-imagenetViTAE-H + MAE (448)#47Top 1 Accuracy: 88.5%Number of params: 644M
instance-segmentation-on-coco-minivalViTAEv2-S + Cascade Mask R-CNN#66mask AP: 44.5
semantic-segmentation-on-ade20kViTAEv2-S + UperNet (multi-scale)#170Validation mIoU: 48.0
object-detection-on-coco-val2017ViTAEv2-S + Cascade Mask R-CNN#188box AP: 51.4

Rank counts only results with a code link.