A paper titled Hybrid Mesh-Neural Representation for 3D Transparent Object Reconstruction, co-authored by Professors Hujun Bao and Weiwei Xu of Zhejiang University's CAD&CG State Key Laboratory, Dr. Jiamin Xu of the iGames team at Hangzhou Dianzi University, and Ph.D. student Zihan Zhu of ETH Zurich, has received the 2025 Best Paper Nomination Award from the journal Computational Visual Media(CVMJ).
The paper introduces a 3D reconstruction method for transparent objects based on a hybrid mesh-neural representation. Its core innovations include:
1. Hybrid geometric representation: combining explicit meshes with local MLPs to accurately capture the fine surface structures of transparent objects. A set of small-scale MLP networks is used to speed up the convergence of the optimization algorithm and enhance reconstruction quality.
2. Ray-cell correspondence: proposing a more accessible light-path constraint that reduces the complexity of implementing Environment Matting and simplifies the capture process under natural illumination.
Experimental results demonstrate that the method can produce high-quality 3D reconstructions of transparent objects from images taken with handheld devices under natural lighting conditions.
Figure 1: Reconstruction pipeline for transparent objects
Figure 2: Reconstruction and re-rendering results of transparent objects
Computational Visual Media was founded in 2015 as the first English-language scientific journal dedicated to visual media research in China and Asia. It is sponsored by Tsinghua University, published by Tsinghua University Press, and academically supported by the Graphics Laboratory of the Department of Computer Science and Technology at Tsinghua University. The editor-in-chief is Professor Shi-Min Hu, an academician of the Chinese Academy of Sciences. The journal publishes high-quality original research and review articles covering the theory, methods, and applications of intelligent visual media processing, and actively encourages interdisciplinary research spanning computer graphics, computer vision, image and video processing, geometric computing, and artificial intelligence. It aims to broaden knowledge horizons and research perspectives for visual media researchers both in China and internationally, serving as a bridge and platform for academic exchange. The journal is indexed in major databases including SCIE, Scopus, and EI. Its latest impact factor, announced in June 2025, is 18.3, and it has ranked No. 1 for two consecutive years among the 129 SCIE/ESCI journals in Clarivate's Journal Citation Reports category of “Computer Science, Software Engineering.” In November 2024, it was selected for the second phase of the Excellence Action Plan for China's Scientific Journals as a leading English-language journal. In March 2026, it was included in the China Computer Federation (CCF) Recommended List of International Conferences and Journals as a Class B journal in the Computer Graphics and Multimedia category.
Dr. Jiamin Xu of the iGames team at Hangzhou Dianzi University is the first author of the paper. The research was carried out in collaboration with Zihan Zhu during their time at the CAD&CG State Key Laboratory of Zhejiang University, under the joint supervision of Professors Hujun Bao and Weiwei Xu. Professor Xu served as the corresponding author. The code is open-sourced and available at: https://github.com/superxjm/HybridTransparentRecon
