Title: Semantic Segmentation of Clothes in the Context of Soft Biometrics Using Deep Learning Methods
Authors: Andrei de Souza Inácio, Anderson Brilhador, Heitor Silvério Lopes
Abstract: Soft biometrics is an emerging area of research, mainly due to its large applicability in people surveillance. It is related to human characteristics that can be used for people classification based on appearance, including: physical, behavioral or adhered (such as clothing) features. Semantic segmentation of clothes is still a challenge for researchers because of the wide variety of clothing styles, layering, and shapes. This work presents an approach for clothing semantic segmentation tasks using the Feature Pyramid Network (FPN) with the EfficientNet as the backbone. We compare this approach with three other deep learning architectures: LinkNet, PSPNet, and U-Net. Due to the lack of a large dataset to train the deep learning model, we propose a combination of two datasets: CCP and CFPD, with refined labels to reduce similar classes. The resulting dataset contains 3,686 images with pixel-level annotations in 15 different categories. Experimental results show the effectiveness of our approach.
Key-words: Clothing Segmentation; Deep Learning; Clothing Parsing; Soft Biometrics
DOI code: 10.21528/CBIC2019-77
PDF file: CBIC2019-77.pdf
BibTeX file: CBIC2019-77.bib