YoloV5 improvement strategy: AAAI 2024 latest axial attention | Plug and play, the first choice for improvement | First launched on the entire network, including data sets and codes, ready to use out of the box!

Summary

https://arxiv.org/pdf/2312.08866.pdf
This paper proposes a method called Multi-scale Cross-axis Attention (MCA) to solve the problem of multi-scale information and long-distance dependence in medical image segmentation . This method is based on efficient axial attention and better captures global information by calculating the bidirectional cross-attention between two parallel axial attentions. To handle significant variations in individual sizes and shapes of diseased regions or organs, we also use multiple bar convolutions with different kernel sizes in each axial attention path to improve the efficiency of encoding spatial information. We build the proposed MCA on the MSCAN backbone to form a network named MCANet. Using only 4M+ parameters, our MCANet outperforms most previous work using heavy-duty backbones on four challenging tasks, including skin lesion segmentation, nuclear segmentation, abdominal multi-organ segmentation and polyp segmentation. This method can be used for medical image segmentation, helping doctors conduct diagnosis and pathological research, and improving the accuracy of diagnosis.

YoloV5 official test results

   YOLOv5l summary: 267 layers, 46275213 parameters, 0 gradients