MHCP-RCNN: Multi-Human Color Parsing Segmentation using Multi-Task Network

Instance and semantic level segmentation are the two most widely used research topics used with assorted and diverse applications. Even though recent advancements in object detection and segmentation have grown by leaps and bounds, using it to better understand the environment and the people in it with more detail, instance-level segmentation and human parsing techniques are used which is still an essential provision. The parsing-by-detection multistage pipeline is used by other comparable research which objectively relies on separately trained detection and models for segmenting data to identify occurrences, before sequentially performing human parsing for each instance. Hence in this paper we propose a novel multi-task multi-stage network that incorporates human detection with the segmentation of body parts along with mapping its semantic color features for multiple person in a single pass. To benchmark, we have used a recently launched CCIHP dataset that consists of semantic body parts with its legion colour maps, size and patterns where color map extraction is important for fine-grained segmentation of a person's clothes which is the novel and cardinal aspect in the proposed architecture. The model is evaluated on standard datasets and have observed to achieve an increase of 12% in parsing and 3% in semantic color map mIOU scores which is an exceptionally unrivaled metric value comparing to the existing state of the art (SOTA) models.