In this work, we propose a novel approach with neural rendering (RC-MVSNet) to solve such ambiguity issues of correspondences among views. Specifically, we impose a depth rendering consistency loss to constrain the geometry features close to the object surface to alleviate occlusions. Concurrently, ...
git clone https://github.com/Boese0601/RC-MVSNet.git cd RC-MVSNet The code is tested with Python == 3.7, PyTorch == 1.10.1 and CUDA == 11.3 on NVIDIA GeForce RTX 3090. We recommend you to useanacondato manage dependencies. You may need to change the torch and cuda version in the...