MaxViT: Multi-Axis Vision Transformer

Open paper
Benchmark Model Rank Results
image-classification-on-imagenetMaxViT-XL (512res, JFT)#20Top 1 Accuracy: 89.53%
image-classification-on-imagenetMaxViT-L (512res, JFT)#23Top 1 Accuracy: 89.41%
image-classification-on-imagenetMaxViT-XL (384res, JFT)#24Top 1 Accuracy: 89.36%
image-classification-on-imagenetMaxViT-L (384res, JFT)#26Top 1 Accuracy: 89.12%
image-classification-on-imagenetMaxViT-B (512res, JFT)#33Top 1 Accuracy: 88.82%
image-classification-on-imagenetMaxViT-XL (512res, 21K)#36Top 1 Accuracy: 88.7%
image-classification-on-imagenetMaxViT-B (384res, JFT)#37Top 1 Accuracy: 88.69%
image-classification-on-imagenetMaxViT-XL (384res, 21K)#45Top 1 Accuracy: 88.51%
image-classification-on-imagenetMaxViT-L (512res, 21K)#50Top 1 Accuracy: 88.46%
image-classification-on-imagenetMaxViT-B (512res, 21K)#54Top 1 Accuracy: 88.38%
image-classification-on-imagenetMaxViT-L (384res, 21K)#58Top 1 Accuracy: 88.32%
image-classification-on-imagenetMaxViT-B (512res)#135Top 1 Accuracy: 86.7%
image-classification-on-imagenetMaxViT-L (384res)#159Top 1 Accuracy: 86.4%
image-classification-on-imagenetMaxViT-B (384res)#164Top 1 Accuracy: 86.34%
image-classification-on-imagenetMaxViT-S (512res)#183Top 1 Accuracy: 86.19%
image-classification-on-imagenetMaxViT-T (384res)#215Top 1 Accuracy: 85.72%
image-classification-on-imagenetMaxViT-L (224res)#269Top 1 Accuracy: 85.17%Number of params: 212MGFLOPs: 43.9
image-classification-on-imagenetMaxViT-B (224res)#291Top 1 Accuracy: 84.94%Number of params: 120MGFLOPs: 23.4
image-classification-on-imagenetMaxViT-S (224res)#340Top 1 Accuracy: 84.45%Number of params: 69MGFLOPs: 11.7
image-classification-on-imagenetMaxViT-T (224res)#435Top 1 Accuracy: 83.62%Number of params: 31MGFLOPs: 5.6
object-detection-on-coco-val2017MaxViT-B (Cascade Mask R-CNN)#142box AP: 53.4AP50: 72.9AP75: 58.1Params (M): 157
object-detection-on-coco-val2017MaxViT-S (Cascade Mask R-CNN)#150box AP: 53.1AP50: 72.5AP75: 58.1Params (M): 107
object-detection-on-coco-val2017MaxViT-T (Cascade Mask R-CNN)#175box AP: 52.1AP50: 71.9AP75: 56.8Params (M): 69

Rank counts only results with a code link.