DCSP-RCNN: A Network for Person Detection with Color, Size and Pattern Characteristic Parsing
S. K. Abhilash, Nischal DS, Shiv Kumar, Venu Madhav Nookala, S. Sam Karthik · 2023
Traditional human parsing models rely on anchor boxes and do not analyze pixels at the pixel level. This has the potential to restrict their performance. To solve these short-comings, a new model known as DCSP-RCNN( A Network for Person Detection with Color, Size and Pattern Characteristic Parsing) was developed. It analyses pixels at a fine-grained level using two subnetworks, a detection head and an edge-guided parsing module. This allows it to outperform earlier models. DCSP-RCNN additionally employs a refinement head, as well as distinct colour, size, and pattern characteristics, to improve the quality of the semantic parsing outputs. This is accomplished by combining an innovative loss function with both bounding box-level and part-level semantic parsing quality. Experiments on multiple human parsing datasets (CCIHP) reveal that DCSP-RCNN outperforms the best current architectures. A Network for Person Detection with Color, Size and Pattern Characteristic Parsing