FASHION COMPATIBILITY PREDICTION BASED ON MULTI-LAYER MASK TRANSFORMER

In order to solve the problem that the clothing compatibility method cannot fuse the complex relationship features among the items inside the suit well, resulting in low accuracy, this paper proposes a multi-layer mask Transformer model (MLMT) to solve the clothing compatibility problem. We proposed a Transformer-based encoder to fuse the style information of all items inside the suit, and we proposed a mask model to determine the compatibility of garments by comparing the correlation between items. It is validated on the polyvore-T public dataset, and the effectiveness of this method is demonstrated by comparing it with existing methods through experiments.