Describing clothing in human images: A parsing-pose integrated approach

With the advent of information technology, digital product information grows exponentially. People are exposed to far too much information, and information overload can slow down, instead of speeding up, a simple decision-making process like searching for suitable clothing online. Traditional semantic-based product retrieval may not be effective due to human subjectivity and cognitive differences. In this paper, we propose a method by integrating the state-of-the-arts deep neural models in pose estimation, human parsing and category classification to recognise from human images all clothing items and their fine-grained product category information. The proposed fine-grained clothing classification model can facilitate a wide range of applications such as the automatic annotation of clothing images. The effectiveness of the proposed method is validated through experiment on a real-world dataset.