Open halil-karabacak opened 1 year ago
Unfortunately the algorithm only works for a moving camera pointing at static scenes. Tracking heavily relies on this assumption, and if it is violated, failure is inevitable.
The meaning of depth_scale
is the unit of the depth image. Usually it is 1000.0, meaning that the pixel values are measured in millimeters, and depth_scale
converts it to meters.
Thanks @theNded ! I wonder if there is workaround of this problem? Like maybe passing identity matrix to tracking etc. but this time I think tsdf would blow up, but just to confirm.
I think your best shot would be https://github.com/isl-org/Open3D/blob/master/examples/python/t_reconstruction_system/integrate.py. You can fabricate a path_trajectory
file with all identity poses, or simply replace extrinsics
in this line to be identity: https://github.com/isl-org/Open3D/blob/master/examples/python/t_reconstruction_system/integrate.py#L53
Thank you @theNded. I will try it. And do I need to have poses of the camera for each frame if I moved to a portable camera? I guess I will only have depth maps and RGB images, idk even if this is doable. I mean if I need to have poses (.log) files than probably I can't use it because I will have a high FPS ~ 500/600FPS low ROI camera
@halil-karabacak can you please share the command you used to compile with CUDA on Windows?
@halil-karabacak can you please share the command you used to compile with CUDA on Windows?
@FraDePace I don't exactly remember the commands but I build from source as explained in the official doc https://www.open3d.org/docs/latest/compilation.html
Checklist
master
branch).My Question
Hi. I am using open3d 0.17 with python 3.9 using cuda on Windows 10. I can succesfully run the example with the LoungeRGBDImages dataset. My problem is currently I don't have a portable camera, so I tried to copy the pipeline by rotating the object instead of moving the camera and took around 35 frames, which I thought would work. I captured depth maps in 16-bit alongside the rgb images. Finally, I updated the intrinsic matrix and printed it to check if I am going right and printed result is what I expected. I also changed width and height in default_intrinsic.json file. However I am getting
My humble guess is that this error might be related to my stationary camera. However, I am not familiar with the pipeline and this is my first attempt of running & understanding it. I also couldn't understand the meaning of depth_scale Any comment would be appreciated, thanks.