Reference-Free Quality Assessment for Virtual Try-On via Human Feedback

As virtual try-on (VTON) systems become increasingly im-portant in fashion e-commerce, there is a growing need for reliable reference-free evaluation methods, since ground-truth images of the same personwearing the target garment are typically unavailable in real-world scenar-ios. To address this challenge, we propose VTON-IQA, a reference-freeframework for human-aligned image quality assessment without requir-ing ground-truth images. To model human perceptual judgments, we con-struct VTON-QBench, a large-scale human-annotated benchmark com-prising 62,688 try-on images generated by 14 representative VTON mod-els and 431,800 quality annotations collected from 13,838 qualified anno-tators. To the best of our knowledge, this is the largest dataset to date forhuman subjective evaluation in VTON. Extensive experiments show thatVTON-IQA achieves reliable human-aligned image quality assessment.Moreover, we conduct a comprehensive benchmark evaluation of 14 rep-resentative VTON models using VTON-IQA. The dataset and sourcecode are released at https://github.com/litelightlite/VTON-IQA.