CVPR

LiteMVS: Efficient Multi-View Stereo with Foundation Distillation and Expert Aggregation

LiteMVS：结合基础模型蒸馏与专家聚合的高效多视图立体匹配

Tianbao Zhang, Zeyu Liu, Shuyu Wu, Fanxing Li, Zhaoxin Fan, Wenjun Wu, Danping Zou

AI总结 LiteMVS是集成平面扫描几何推理与单目先验的轻量多视图深度模型，通过MoE和基础模型蒸馏提升重建质量与效率，在ScanNetv2等数据集上表现优异。

Comments CVPR 2026 Workshop accepted