Open TaoshuaiZ opened 1 year ago
This is the output of running other dataset:
And this is the output of running my dataset:
@germanros1987 Related to https://github.com/isl-org/Open3D/discussions/4311#discussioncomment-6916524
In my test, I am working with 2 cameras looking at center of room from opposite corners. I have already done camera pose estimation and the generated 3d pointcloud is proper but quite noisy due to depth errors from realsense.
Thus, I do TSDF integration which works nicely if I let it run for longer, but if I set it to integrate for max 1~2 seconds (vaguely remember, need to reproduce and probably note down pointcloud density w.r.t number of frames used to integrate), the mesh generated is filled with holes.
I faced similar issue, and building a reproducible sample is in my backlog.
@TaoshuaiZ Would you like to post the full log? Empty edges mean the pose estimation is probably wrong; time of Making/Registering Fragments should also be larger than zero.
@saurabheights it would be nice to check the reproducible sample to identify the problem.
@theNded Sorry for the delay. I was able to find the issue with your help in issue #4469 . Setting weight threshold to -1 allowed me to get a dense mesh easily.
@TaoshuaiZ , have you found a solution to your problem? I’m facing the same issue. Thanks
Checklist
master
branch).My Question
I want to do 3D reconstruction using RGB-D images.I followed the document Docs>Rconstruction System>System Overview The RGB-D datasets that I provided contains 10 RGB-D pairs from multi-view. And the format of my dataset is the same as the example provided in the document. But when I finish my process of running run_system.py, I can't get my .ply mesh, it says [Open3D WARNING] Write PLY failed: mesh has 0 vertices. and [Open3D WARNING] PoseGraph read JSON failed: empty edges.. What should I do to solve this problem?Should I provide more images?