TY - GEN
T1 - FashionAI
T2 - 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019
AU - Zou, Xingxing
AU - Kong, Xiangheng
AU - Wong, Waikeung
AU - Wang, Congde
AU - Liu, Yuguang
AU - Cao, Yang
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/6
Y1 - 2019/6
N2 - Fine-grained attribute recognition is critical for fashion understanding, yet is missing in existing professional and comprehensive fashion datasets. In this paper, we present a large scale attribute dataset with manual annotation in high quality. To this end, complex fashion knowledge is disassembled into mutually exclusive concepts and form a hierarchical structure to describe the cognitive process. Such well-structured knowledge is reflected by dataset in terms of its clear definition and precise annotation. The problems which are common in the process of annotation, including structured noise, occlusion, uncertain problems, and attribute inconsistency, are well addressed instead of merely discarding those bad data. Further, we propose an iterative process of building a dataset with practical usefulness. With 24 key points, 245 labels that cover 6 categories of women's clothing, and a total of 41 subcategories, the creation of our dataset drew upon a large amount of crowd staff engagement. Extensive experiments quantitatively and qualitatively demonstrate its effectiveness.
AB - Fine-grained attribute recognition is critical for fashion understanding, yet is missing in existing professional and comprehensive fashion datasets. In this paper, we present a large scale attribute dataset with manual annotation in high quality. To this end, complex fashion knowledge is disassembled into mutually exclusive concepts and form a hierarchical structure to describe the cognitive process. Such well-structured knowledge is reflected by dataset in terms of its clear definition and precise annotation. The problems which are common in the process of annotation, including structured noise, occlusion, uncertain problems, and attribute inconsistency, are well addressed instead of merely discarding those bad data. Further, we propose an iterative process of building a dataset with practical usefulness. With 24 key points, 245 labels that cover 6 categories of women's clothing, and a total of 41 subcategories, the creation of our dataset drew upon a large amount of crowd staff engagement. Extensive experiments quantitatively and qualitatively demonstrate its effectiveness.
UR - https://www.scopus.com/pages/publications/85083339753
U2 - 10.1109/CVPRW.2019.00039
DO - 10.1109/CVPRW.2019.00039
M3 - Conference article published in proceeding or book
AN - SCOPUS:85083339753
T3 - IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops
SP - 296
EP - 304
BT - Proceedings - 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019
PB - IEEE Computer Society
Y2 - 16 June 2019 through 20 June 2019
ER -