BiFormer: Vision Transformer with Bi-Level Routing Attention

Open paper
Benchmark Model Rank Results
semantic-segmentation-on-ade20kBiFormer-B (IN1k pretrain, Upernet 160k)#101Validation mIoU: 51.7
semantic-segmentation-on-ade20kUpernet-BiFormer-S (IN1k pretrain, Upernet 160k)#120Validation mIoU: 50.8
image-classification-on-imagenetBiFormer-B* (IN1k ptretrain)#249Top 1 Accuracy: 85.4%
image-classification-on-imagenetBiFormer-S* (IN1k ptretrain)#357Top 1 Accuracy: 84.3%
image-classification-on-imagenetBiFormer-T (IN1k ptretrain)#683Top 1 Accuracy: 81.4%
object-detection-on-coco-val2017BiFormer-B (Mask R-CNN 1x)#252box AP: 48.6AP50: 70.5AP75: 53.8
object-detection-on-coco-val2017BiFormer-S (Mask R-CNN 1x)#266box AP: 47.8AP50: 69.8AP75: 52.3

Rank counts only results with a code link.