Title:	AUTOMATIC LABELING FOR FASHION DATASETS
Author(s):	Anissa Selmani, Houda Bakir and Hedi Zaher
ISBN:	978-989-8704-21-4
Editors:	Yingcai Xiao, Ajith P. Abraham and Jörg Roth
Year:	2020
Edition:	Single
Keywords:	Labelling, Segmentation, Human Pose, Fuzzy Logic
Type:	Full
First Page:	107
Last Page:	114
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Creating image Datasets is a time-consuming step and a challenge for specific classification problems. The dilemma of this task concerns the manual annotation of training images. Thus, this paper presents an automatic labeling method without humans in the loop for Fashion datasets. First, we introduce a new descriptor for texture segmentation called LNLBP (Large Neighbourhood Local Binary Pattern) which is able to capture both micro-structure and macro-structure texture information. Then, using a Fuzzy logic system, the proposed method detects the change in texture, intensity and human pose in order to localize correctly the position of the fashion item. Finally, we generate the ‘xml’ files containing the coordinates and the labels in order to use them afterward for the training with the Google Tensorflow object detection API: for real-time clothes classification. The resulting dataset will be made publicly available for research.
	Go Back