DKTNet: Dual-Key Transformer Network for small object detection

Shoukun Xu, Jianan Gu, Yining Hua, Yi Liu

Abstract: Highlights•Dual-Key Transformer Network is proposed for small object detection in complex backgrounds.•2D convolution computes Q, Kdual<math><mrow is="true"><msub is="true"><mrow is="true"><mi is="true">K</mi></mrow><mrow is="true"><mi mathvariant="italic" is="true">dual</mi></mrow></msub></mrow></math> and V to preserve the local context for the transformer.•1D convolution is developed to further enhance computational efficiency.•The framework achieves the state-of-the-art performance.