Abstract

Abstract: A machine must be able to detect individual object instances as well as how they interact in order to comprehend thevisual environment. Humans are frequently at the heart of such relationships, and recognizing such interactions is a significantpractical and scientific challenge. For human part detection, there is a dearth of a large-scale, well-annotated dataset. COCOHuman Parts is a solution that fills the void. The proposed data set is based on COCO 2017, the first instance-level human parts data set, and features photos of complex scenarios with a wide range of variation. Previous dataset is the goal of advancing thestate-of-the-art in object recognition by placing the question of object recognition in the context of the broader question of scene understanding. It provides baseline performance analysis for bounding box and segmentation detection results using a Deformable Parts Model. For detection of human parts, we develop a new dataset Hier R-CNN. In this dataset we propose a strong baselinefor detecting human parts at instance-level over this data set in an end-to-end manner. R-CNN detects human parts of each person and predict the subordinate relationship between them.Keywords— COCO human parts, region-based method, Hier R-CNN, Object detection.