Focal Modulation Networks

Benchmark Model Rank Results
object-detection-on-coco-test-devFocalNet-H (DINO)#8box mAP: 64.4
object-detection-on-coco-val2017FocalNet-H (DINO)#10box AP: 64.2
object-detection-on-coco-val2017FocalNet-T (LRF, Cascade Mask R-CNN)#186box AP: 51.5AP50: 70.3AP75: 56.0
object-detection-on-coco-val2017FocalNet-T (SRF, Cascade Mask R-CNN)#447AP50: 70.1AP75: 55.8
panoptic-segmentation-on-coco-minivalFocalNet-L (Mask2Former (200 queries))#16PQ: 57.9AP: 48.4
semantic-segmentation-on-ade20kFocalNet-L (Mask2Former)#21Validation mIoU: 58.5
instance-segmentation-on-coco-minivalFocalNet-B + Mask R-CNN#72mask AP: 44.1
image-classification-on-imagenetFocalNet-L#105Top 1 Accuracy: 87.3%