C2CNet: Connect to Casual Feature Pyramid Net for Target Detection

Currently, the existing target detection methods have many limitations such as low accuracy, high false detection rate, etc. To solve these problems, this paper proposed a C2CNet, named Connect to Casual feature pyramid Net, which aims to improve the accuracy of target detection. First, the C2C module is designed to increase the network's capability of information fusion and improve the model's detection accuracy for objects of different sizes through a large number of residual structures. Secondly, the commonly used SPPF (namely Spatial Pyramid Pooling-Fast) part in YOLOv5 is replaced by SPPFCSPC+ (namely Spatial Pyramid Pooling-Fully Connnected Spatial Pyramid Convolution) with serialization to deepen the utilization of feature information. We evaluated C2CNet on the VOC2012 and MSCOCO datasets. The results showed that the detection [email protected] of C2CNet reached 78.7% on VOC2012 and 70.6% on MSCOCO. The detection accuracy is also significantly improved compared to other related approaches (such as Swin and YOLOv7).