Conv and Efficient Multi-Scale Attention Module for YOLOv5

Object detection is a popular and tough task, the multiplicity of species and the multi-scales of the same kind contribute to these difficulties, as a result, the performance improvement of various algorithms is minimal. The proposed attention mechanism can improve the performance of the object detection algorithm. However, embedding the attention mechanism may bring negative benefits to the algorithm model, because of the depth of the algorithm model, the attention mechanism can be embedded in many different locations, and different locations will bring different results. To improve the object detection performance of YOLOv5, this paper proposes Conv and Efficient Multi-Scale Attention (CEMA), a new novel module used in YOLOv5, it fusion C3 module and EMA attention. The performance at different locations is compared and analyzed. Experimental results used the VOC2007 + 2012 dataset to verify its performance. It has a 1.76% improvement compared with YOLOv5s, and it also has different degrees of improvement compared with SE, CA, CBAM, and SimAM attention mechanisms.