Title:	DETECTION OF FASHION LANDMARKS BASED ON POSE ESTIMATION AND HUMAN PARSING
Author(s):	Honghong He, Yanghong Zhou, Jin-tu Fan and P. Y. Mok
ISBN:	978-989-8704-42-9
Editors:	Yingcai Xiao, Ajith Abraham, Guo Chao Peng and Jörg Roth
Year:	2022
Edition:	Single
Keywords:	Fashion Landmarks Localization, Pose Estimation, Human Parsing, Graph-based Clothing Structure
Type:	Full Paper
First Page:	62
Last Page:	69
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Locating fashion items on any input images is referred as 'fashion image understanding', which is often a preliminary or initial step that supports various other visions tasks. The study of fashion image understanding has been widely benefited from the rapid advancements in deep learning-based new models and the availability of large datasets covering real-world fashion images. However, the real value of these datasets is in doubt, because different annotation strategies have been adopted in these datasets, resulting in different landmarks for different clothing styles and low compatibility of these datasets. In this paper, we propose a pose-aware segmentation-based method to locate key points of fashion items on fashion images by taking advantage of clear correspondence between clothing and human body, that can be applied to locate key points of fashion items and to re-annotate existing datasets for cross dataset learning. The validity of the method was validated on a subset of the DeepFashion2 dataset.
	Go Back