Hi3DEval: Advancing 3D Generation Evaluation with Hierarchical Validity

Overall framework

Abstract

Hi3DEval is a hierarchical framework for evaluating generated 3D assets. It combines object-level and part-level assessment, extends texture evaluation toward material realism, introduces Hi3DBench, and proposes a 3D-aware automated scoring system aligned with human preference.

Publication
In Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track
Tong WU 吴桐
Tong WU 吴桐
Assistant Professor @ Fudan

My research interests include 3d vision, long-tailed recognition, and robustness.

Related