3D MeanFlow: One-Step Point Cloud Completion and Generation via Average-Velocity Transport

Haowen Zhong, Jiujun Cheng, Haowen Wang, Chao Wei, Lu Yang, Qichao Mao, Shangce Gao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:164235-164263, 2026.

Abstract

Point cloud completion and generation are important across many 3D tasks, where both fidelity and sampling efficiency matter. Prevailing high-fidelity approaches rely on long sampling schedules, which incur substantial inference latency. Few-step alternatives typically use rectification or distillation, leading to multi-stage training pipelines and potential quality trade-offs. We present 3D MeanFlow (3DMF), a distillation-free model that performs one-step average-velocity transport for point cloud completion and generation. We optimize an instantaneous-average consistency objective and impose a shape-level constraint to stabilize training. Additionally, we introduce PointPlug, integrating completion into 3D object detectors and evaluating its impact. PointPlug uses adaptive selection that balances benefit and latency. Across standard benchmarks, 3DMF achieves one-step sampling with an order-of-magnitude speedup while maintaining competitive fidelity. On nuScenes and KITTI, inserting PointPlug improves all evaluated detectors under comparable settings.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-zhong26b, title = {3{D} {M}ean{F}low: One-Step Point Cloud Completion and Generation via Average-Velocity Transport}, author = {Zhong, Haowen and Cheng, Jiujun and Wang, Haowen and Wei, Chao and Yang, Lu and Mao, Qichao and Gao, Shangce}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {164235--164263}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhong26b/zhong26b.pdf}, url = {https://proceedings.mlr.press/v306/zhong26b.html}, abstract = {Point cloud completion and generation are important across many 3D tasks, where both fidelity and sampling efficiency matter. Prevailing high-fidelity approaches rely on long sampling schedules, which incur substantial inference latency. Few-step alternatives typically use rectification or distillation, leading to multi-stage training pipelines and potential quality trade-offs. We present 3D MeanFlow (3DMF), a distillation-free model that performs one-step average-velocity transport for point cloud completion and generation. We optimize an instantaneous-average consistency objective and impose a shape-level constraint to stabilize training. Additionally, we introduce PointPlug, integrating completion into 3D object detectors and evaluating its impact. PointPlug uses adaptive selection that balances benefit and latency. Across standard benchmarks, 3DMF achieves one-step sampling with an order-of-magnitude speedup while maintaining competitive fidelity. On nuScenes and KITTI, inserting PointPlug improves all evaluated detectors under comparable settings.} }
Endnote
%0 Conference Paper %T 3D MeanFlow: One-Step Point Cloud Completion and Generation via Average-Velocity Transport %A Haowen Zhong %A Jiujun Cheng %A Haowen Wang %A Chao Wei %A Lu Yang %A Qichao Mao %A Shangce Gao %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhong26b %I PMLR %P 164235--164263 %U https://proceedings.mlr.press/v306/zhong26b.html %V 306 %X Point cloud completion and generation are important across many 3D tasks, where both fidelity and sampling efficiency matter. Prevailing high-fidelity approaches rely on long sampling schedules, which incur substantial inference latency. Few-step alternatives typically use rectification or distillation, leading to multi-stage training pipelines and potential quality trade-offs. We present 3D MeanFlow (3DMF), a distillation-free model that performs one-step average-velocity transport for point cloud completion and generation. We optimize an instantaneous-average consistency objective and impose a shape-level constraint to stabilize training. Additionally, we introduce PointPlug, integrating completion into 3D object detectors and evaluating its impact. PointPlug uses adaptive selection that balances benefit and latency. Across standard benchmarks, 3DMF achieves one-step sampling with an order-of-magnitude speedup while maintaining competitive fidelity. On nuScenes and KITTI, inserting PointPlug improves all evaluated detectors under comparable settings.
APA
Zhong, H., Cheng, J., Wang, H., Wei, C., Yang, L., Mao, Q. & Gao, S.. (2026). 3D MeanFlow: One-Step Point Cloud Completion and Generation via Average-Velocity Transport. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:164235-164263 Available from https://proceedings.mlr.press/v306/zhong26b.html.

Related Material