A quick hack to combine DSNeRF with R3LIVE.
The idea of the project is to provide depth-supervision from dense point clouds for a NeRF.
This is the DSNerf tutorial about how to use depth-supervison(sparse point cloud using COLMAP).
Use the files in rosbagUtils/
to extract data from a ROS bag. Details sees there
An example R3LIVE bag is Here. It is a 5 senconds segment from R3LIVE dataset with all the ouputs from R3LIVE.
Create a dataset under data/<exp_name>
. After extract data from rosbag, we should have <img_dir>, <points_dir>, <poses_dir>
. Put the images, points and poses(odom) files into the folder like below.
├── data
│ └── hkust_2view_ros
│ ├── images
│ │ ├── frame000001.jpg
│ │ └── frame000075.jpg
│ ├── ros
│ │ ├── odom
│ │ │ ├── 0001.npy
│ │ │ └── 0075.npy
│ │ └── points
│ │ ├── 0001.npy
│ │ └── 0051.npy
NOTE
Because the LiDAR scanned points and images captured by the camera are not synchronized, we need to manually select the proper data frame according to the timestamp provided in timestamp.txt
.
python ros2poses.py data/<exp_name>
It will create a file poses_bounds.npy
under data/<exp_name>
which can be loaded as llff datatype.
In config/<config_file>
, be sure to set ros_depth=True
, then run
python run_nerf.py --config config/<config_file>