Separable Self-attention for Mobile Vision Transformers

Benchmark Model Rank Results
image-classification-on-imagenetMobileViTv2-1.0#802Top 1 Accuracy: 78.1%Number of params: 4.9MGFLOPs: 1.8
image-classification-on-imagenetMobileViTv2-0.75#883Top 1 Accuracy: 75.6%Number of params: 2.9MGFLOPs: 1.0
image-classification-on-imagenetMobileViTv2-0.5#958Top 1 Accuracy: 70.2%Number of params: 1.4MGFLOPs: 0.5