In virtual try-on networks,the segmentation determines the accuracy of apparel mi-gration and transformation from the model to the target skeleton structure.An energy-based spatial attention algorithm to virtual try-on application is proposed for obtaining more accurate segmentation encoding results within dressing-fitting-like generation networks.This algorithm uses the projected energy of the target subject on different axis directions to form a focus area of spatial position,providing effective spatial attention pre-processing information for segmenta-tion encoder.With experimental simulations,the proposed method can achieve a good improve-ment in segmentation encoding.
维普
万方数据