Details

Autor(en) / Beteiligte

• Deng, Hanhui
• Jiang, Jianan
• Yu, Zhiwang
• Ouyang, Jinhui
• Wu, Di

Titel

CrossGAI: A Cross-Device Generative AI Framework for Collaborative Fashion Design

Ist Teil von

• Proceedings of ACM on interactive, mobile, wearable and ubiquitous technologies, 2024-03, Vol.8 (1), p.1-27, Article 35

Ort / Verlag

New York, NY, USA: ACM

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Fashion design usually requires multiple designers to discuss and collaborate to complete a set of fashion designs, and the efficiency of the sketching process is another challenge for personalized design. In this paper, we introduce a fashion design system, CrossGAI, that can support multiple designers to collaborate on different devices and provide AI-enhanced sketching assistance. Based on the design requirements analysis acquired from the formative study of designers, we develop the system framework of CrossGAI implemented by the user-side web-based cross-device design platform working along with the server-side AI-integrated backend system. The CrossGAI system can be agilely deployed in LAN networks which protects the privacy and security of user data. To further improve both the efficiency and the quality of the sketch process, we devised and exploited generative AI modules, including a sketch retrieval module to retrieve sketches according to stroke or sketch drawn, a sketch generation module enabling the generation of fashion sketches consistent with the designer's unique aesthetic, and an image synthesis module that could achieve sketch-to-image synthesis in accordance with the reference image's style. To optimise the computation offloading when multiple user processes are handled in LAN networks, Lyapunov algorithm with DNN actor is utilized to dynamically optimize the network bandwidth of different clients based on their access history to the application and reduce network latency. The performance of our modules is verified through a series of evaluations under LAN environment, which prove that our CrossGAI system owns competitive ability in AIGC-aided designing. Furthermore, the qualitative analysis on user experience and work quality demonstrates the efficiency and effectiveness of CrossGAI system in design work.