June 20, 2026

Accepted to ECCV 2026 — Code, Pretrained Models, and Dataset Now Public (Joint Work with ZOZO NEXT)

Our joint project with ZOZO NEXT has been accepted to
ECCV 2026 (European Conference on Computer Vision),
a top-tier international conference in computer vision.

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

Authors
Yuki Hirakawa, Takashi Wada, Ryotaro Shimizu, Takuya Furusawa, Yuki Saito,
Ryosuke Araki, Tianwei Chen, Fan Mo, Yoshimitsu Aoki

Project Lead
Yuki Hirakawa (D1)

Overview
We propose VTON-IQA, a new reference-free image quality assessment (IQA)
method that evaluates the quality of virtual try-on images without requiring a reference image.
We also build VTON-QBench, a large-scale benchmark consisting of 62.7K try-on
images and 431.8K human annotations. The code, pretrained models, and VTON-QBench are now
publicly available, and an interactive demo can be tried on Hugging Face.

• Paper (arXiv):
  arxiv.org/abs/2603.13057
• Code & Models (GitHub):
  github.com/litelightlite/VTON-IQA
• Demo (Hugging Face):
  huggingface.co/spaces/zozonext/VTON-IQA