PGNet: Pipeline Guidance for Human Key-Point Detection.

摘要: Human key-point detection is a challenging research field in computer vision. Convolutional neural models limit the number of parameters and mine local structure, have made great progress significant target detection. However, features extracted by shallow layers mainly contain lack semantic information, while deep rich information but spatial that results imbalance feature extraction imbalance. With complexity network structure increasing amount computation, balance between time communication calculation highlights importance. Based on improvement hardware equipment, operation greatly improved optimizing data methods. as becomes deeper deeper, consumption networks also increases, computing capacity optimized. In addition, overhead focus recent attention. We propose novel PGNet, which contains three parts: pipeline guidance strategy (PGS); Cross-Distance-IoU Loss (CIoU); Cascaded Fusion Feature Model (CFFM).