SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications

Open paper
Benchmark Model Rank Results
instance-segmentation-on-coco-minivalSwiftFormer-L3 + Mask R-CNN#96mask AP: 39.1
instance-segmentation-on-coco-minivalSwiftFormer-L1 + Mask R-CNN#101mask AP: 38.1
semantic-segmentation-on-ade20kSwiftFormer-L3 + Semantic FPN#220Validation mIoU: 43.9
semantic-segmentation-on-ade20kSwiftFormer-L1 + Semantic FPN#230Validation mIoU: 41.4
object-detection-on-coco-val2017SwiftFormer-L3 + Mask R-CNN#355box AP: 42.7
object-detection-on-coco-val2017SwiftFormer-L1 + Mask R-CNN#378box AP: 41.2
image-classification-on-imagenetSwiftFormer-L3#511Top 1 Accuracy: 83.0%
image-classification-on-imagenetSwiftFormer-L1#718Top 1 Accuracy: 80.9%
image-classification-on-imagenetSwiftFormer-S#868Top 1 Accuracy: 78.5%
image-classification-on-imagenetSwiftFormer-XS#979Top 1 Accuracy: 75.7%

Rank counts only results with a code link.