[論文レビュー] Fashion Meets Computer Vision: A Survey

Wen-Huang Cheng, Sijie Song+3 more|arXiv (Cornell University)|Mar 31, 2020

Generative Adversarial Networks and Image Synthesis参考文献 207被引用数 50

ひとこと要約

200を超えるファッション中心のコンピュータビジョン作業の検出、分析、合成、推奨に関する包括的な調査で、データセット、ベンチマーク、今後の方向性を含む。

ABSTRACT

Fashion is the way we present ourselves to the world and has become one of the world's largest industries. Fashion, mainly conveyed by vision, has thus attracted much attention from computer vision researchers in recent years. Given the rapid development, this paper provides a comprehensive survey of more than 200 major fashion-related works covering four main aspects for enabling intelligent fashion: (1) Fashion detection includes landmark detection, fashion parsing, and item retrieval, (2) Fashion analysis contains attribute recognition, style learning, and popularity prediction, (3) Fashion synthesis involves style transfer, pose transformation, and physical simulation, and (4) Fashion recommendation comprises fashion compatibility, outfit matching, and hairstyle suggestion. For each task, the benchmark datasets and the evaluation protocols are summarized. Furthermore, we highlight promising directions for future research.

研究の動機と目的

• Catalog current state-of-the-art in intelligent fashion spanning detection, analysis, synthesis, and recommendation.
• Summarize benchmark datasets and evaluation protocols for each fashion task.
• Provide performance comparisons and insights to guide future research directions.

提案手法

• Organize research into four categories: fashion detection, analysis, synthesis, and recommendation.
• Review landmark detection, parsing, and item retrieval under detection.
• Survey attribute recognition, style learning, and popularity prediction under analysis.
• Summarize style transfer, pose transformation, and physical simulation under synthesis.
• Outline fashion compatibility, outfit matching, and hairstyle suggestion under recommendation.
• Compile datasets and evaluation metrics for each task and highlight datasets links.

実験結果

リサーチクエスチョン

• RQ1 What are the current leading methods and datasets for fashion detection, analysis, synthesis, and recommendation?
• RQ2 What evaluation metrics and benchmarks dominate each fashion task, and how do methods compare?
• RQ3 What are the key challenges and future directions in intelligent fashion research?

主な発見

• Four main research streams define intelligent fashion: detection, analysis, synthesis, and recommendation.
• Deep learning-based approaches dominate landmark detection, parsing, and cross-domain fashion retrieval tasks.
• A wide range of benchmark datasets exist (e.g., DeepFashion, DeepFashion2, LIP, Fashionpedia) with task-specific metrics.
• Survey highlights trends, datasets, and future directions to spur progress in intelligent fashion.
• Performance varies by dataset and task, with newer models generally improving landmark localization and parsing accuracy across benchmarks.