Table 3 Using the DTU dataset for training, we measured the GPU memory required for training the networks of MVSNet and our model, as well as the time needed to predict depth maps for each view.

From: Enhanced multi view 3D reconstruction with improved MVSNet