Blending 3D Geometry and Machine Learning for Multi-View Stereopsis

Vibhas Vats, Md. Alimoor Reza, David Crandall, Soon-heung Jung·May 06, 2025

Summary

GC-MVSNet++ is an advanced algorithm that halves training iterations for multi-view stereo through geometric consistency checks. It excels in multi-view depth estimation, achieving top-tier results on benchmarks like DTU, BlendedMVS, and Tanks and Temples. The method includes a densely connected cost regularization network, enhancing dense feature connections. Improvements over competitors include a GC module with D-FEN modifications, DF preprocessing, and Dense-CostRegNet integration. GC-MVSNet++ achieves state-of-the-art performance, enhancing learning efficiency in multi-view stereo pipelines.

Advanced features