It is an implementation to use MLP decoder to regress the color value using the same structure as the sdf regression in SHINE_MAPPING.For the colored point cloud data,I use R3LIVE to obtain colored point cloud and the sensor pose.And I record rosbag and play it to a script to convert it to kitti format.
-
Run the r3live to obtain the colored point cloud and the sensor pose.If you haven't installed r3live,you can download this rosbag and try it first.rosbag data
-
Run the script to convert the rosbag to kitti format.
python dataset/colorpoints_to_kitti_format.py -
Modify your dataset path in color_incre_reg.yaml in the config directory. And run the shine_incre.py,since the batch is not implemented yet.
python shine_incre.py config/color_incre_reg.yaml
| Raw Pointcloud | Reconstruction Results |
|---|---|
 |
 |
| Raw Pointcloud | Reconstruction Results |
|---|---|
 |
 |
As you can see, the reconstruction quality is not so satisfying and I think there is more can be done to optimize the result.
Thanks to these great works!