[2K5-GS-2] Machine learning

[2K5-GS-2-04] Comments generation using MAGMA for fashion outfit

〇Sayaka Aoba¹, Yugen Sato¹, Ryozo Maskawa¹, Makoto Sato¹, Shosuke Haji¹, Taiga Matsui², Keita Ishikawa², Tomohiro Takagi¹ (1. Meiji University, 2. airCloset, Inc.)

Keywords:Comment generation, Fashion outfit, Lage-scale model, MAGMA, Vision and Language

In this study we build a model to generate comments when recommending two coordinated fashion items in personal styling. In coordinating recommendations, advice is also provided on the compatibility of these items and their combination with accessories and other items. We uses MAGMA, a method that supports multimodal input for language models by using adapters, to generate coordinating comment sentences with the combined image of two clothes and prompts as input. Quantitative and qualitative comparisons with manually generated comments and conventional methods based on machine learning were conducted to confirm that the proposed model is fully capable.