MetaFormer Is Actually What You Need for Vision

Open paper
Benchmark Model Rank Results
image-classification-on-imagenetMetaFormer PoolFormer-M48#515Top 1 Accuracy: 82.5%Number of params: 73MGFLOPs: 23.2
object-detection-on-coco-minivalPoolFormer-S36 (Mask R-CNN)#168box AP: 41.0AP50: 63.1AP75: 44.8
semantic-segmentation-on-ade20kPoolFormer-M48#206Validation mIoU: 42.7
semantic-segmentation-on-densepassPoolFormer (MiT-B1)#9mIoU: 43.18%

Rank counts only results with a code link.