MFAR-VTON: Multi-scale Fabric Adaptive Registration for Image-Based Virtual Try-On

Image-based virtual try-on technology provides a better shopping experience for online consumers. However, existing methods face challenges in effectively capturing high-level semantic information and achieving accurate registration between clothing and the body, particularly in complex body poses or target garments. To address these issues, we propose MFAR-VTON, a novel framework that incorporates a multi-scale enhanced adaptive clothing registration strategy and possesses matching filtering capabilities. Our method enables the generation of highly precise clothing alignment results, leading to seamless integration of try-on images. Additionally, we introduce a deformation energy constraint that effectively preserves intricate garment details. Experimental results demonstrated that MFAR-VTON achieves state-of-the-art performance in terms of accuracy and realism.