Paper Doll Parsing: Retrieving Similar Styles to Parse Clothing Items

Runway to Realway: Visual Analysis of Fashion

“…Recent work addresses attributes of shopping images of clothing items [2,20], adding some clothing attributes to the person detection pipeline [4], and detecting clothing items and annotating attributes for an image or collection of images [6,3]. More closely related to the parsing approach used in this paper, there has been work on predicting semantic segmentations of clothing images [24,25,26], and we use the open source implementation from [26] as part of our pipeline. There is also related work on recognizing attributes or concepts that are strongly related to clothing, for example, occupation [22,21] or social identity [15].…”

“…Typically fine-grained categorization approaches require coarse correspondence of keypoints in order to reliably extract subtle differences in appearance. In this paper, we make use of pose estimation [27] and clothing parsing [26] to automatically localize important regions in a fashion outfit. Using these as an input parse, our approach learns to combine features computed over parsed regions in order to mimic human judgments of fashion similarity.…”

“…Therefore, we develop a feature representation that can usefully capture the appearance of clothing items in outfits. This representation first estimates the pose of a person [27] and what they are wearing in the form of a clothing parse [26], and then computes pose-dependent features related to clothing style. To understand visual relationships between different outfits, we collect human judgments of outfit similarity using Amazon Mechanical Turk.…”

“…Using this dataset we develop computer vision techniques to measure the similarity between images of clothing outfits. Additionally, we make use of the Paper Doll dataset [26] collected from the Chictopia social network, where people interested in fashion post and annotate pictures of their daily outfits. The combination of these two datasets allows us to make initial steps toward analyzing how fashion trends transfer from runway collections to the clothing people wear in real life.…”

Multi-modal Embedding for Main Product Detection in Fashion

“…Common examples are clothing classification or retrieval [3,2,23], clothing parsing (i.e., semantically label each image pixel) [29,21,4,15] or higher level tasks such as evaluating style or deducting people's occupation or social tribe [24,11,28,22,17].…”