JoVox Proposes Lightweight Voxel Generation with Joint Hierarchical Flow Matching
Xiangjun Tang, Jianqi Chen and Peter Wonka submitted the JoVox paper to arXiv on November 7, 2026. It introduces discrete flow matching across all levels of a voxel hierarchy, two-level prediction and an unmasking strategy that revisits earlier decisions. The authors report outperforming baselines, including 8× higher training throughput and one-fifteenth the forward-pass FLOPs of TRELLIS.2. They also report training on 4 rather than 32 GPUs and higher quality, with an 11% lower Chamfer distance; these benchmark results are author-reported.
JoVox presents a computationally efficient approach to generating high-fidelity voxels. The authors report improved training efficiency, lower forward-pass computation and higher quality relative to TRELLIS.2, though the benchmark results have not been independently verified in the supplied report.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.10690 (cs)
[Submitted on 7 Nov 2026]
Title:JoVox: Joint Hierarchical Flow Matching for Lightweight Voxel Generation
Authors:Xiangjun Tang, Jianqi Chen, Peter Wonka
View a PDF of the paper titled JoVox: Joint Hierarchical Flow Matching for Lightweight Voxel Generation, by Xiangjun Tang and 2 other authors