About the parameter tuning of RMVSNet in the provided test images

[sy8008]

Hello! I wonder how did you set the depth related parameters(depth_min, depth_interval, number_depth,etc) when test RMVSNet on the provided test images. It seems that directly use the depth range estimated by colmap will cause problem. I wonder what parameters have you tuned to get a good result when combine colmap with R-MVSNet. I have tried to use the colmap2mvsnet.py provided in RMVSNet's official implementation to convert the colmap SFM result, and then directly feed them into a depth inference network similar to R-MVSNet（named RED-Net）, but the result seems terrible. In RMVSNet's issue, I found many people have encountered such issue when test on custom dataset, it seems that the set of depth range matters a lot for such networks.