Closed ridanlukita closed 2 years ago
Depth unit is hard-coded for sensors. If you are still using RealSense D435i, keep using 1000, and do not change it to 100. Otherwise you need to modify every parameter proportionally, including max_depth
, voxel_size
, max_depth_diff
, tsdf_cubic_size
, etc.
Thank you! Now I able to do that similar to the tutorial
After I read the documentation for each of the parameter you mentioned, I've understand what each of those parameter used for.
Otherwise you need to modify every parameter proportionally, including max_depth, voxel_size, max_depth_diff, tsdf_cubic_size, etc.
But I want to know further, what/how are the rules for combining these parameters to yield a good reconstruction model?
Is there any guide/tutorial for that?
This is not yet available, but we will update that part once the reconstruction system is fully migrated to the tensor interface. I will close this issue for now and feel free to file other issues if necessary.
Checklist
master
branch).My Question
Describe the issue
The reconstruction system works great for near objects, however, if the object further than ±40 cm, the results from integrated.ply (in the folder scene) always just a blank, no objects at all.
I record the RGBD (.jpg color and .png depth) image dataset using Intel RealSense D435i with modified MediumDensityPreset from RealSense Viewer
Steps to reproduce the bug
I had done:
realsense.json configuration
MediumDensityPreset.json
Error message
There is always warning that appear like this in the step (green text) below
Expected behavior
Longer range reconstruction can be generated
Open3D, Python and System information
Additional information
Around 3 months ago everything works normal, I can make 3D reconstruction of a room, similar to http://www.open3d.org/docs/0.13.0/tutorial/reconstruction_system/integrate_scene.html#results
But around last week until today, I often do scan for near object using mini turntable for 360 degree scanning and changing the disparity shift to around 30-50 and use depth units of 100.
Then I use the camera for scanning long range objects, I change the disparity shift to 0 and depth units to 1000. For color and depth dataset from the camera give no problem, but the reconstruction system give me blank results.
Is there any suggestion to solve these problem?