This code is mainly based on DROID-SLAM.
It first estimated camera poses by DROID-SLAM and then predicted dense depth maps by CDS-MVSNet
by using all available pretrained models including droid.pth and cds_mvsnet.pth.
Sampled data and pretrained models can be download here
Run the demo on any of the samples (all demos can be run on a GPU with 8G of memory).
python demo.py --imagedir=data/nmail6 --calib=calib/nmail3.txt --stride=2 --buffer 384 --mvsnet_ckpt cds_mvsnet.pth