search

spla-tam / SplaTAM

SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM (CVPR 2024)
https://spla-tam.github.io/
BSD 3-Clause "New" or "Revised" License
1.58k stars 174 forks source link

Visualization Issues (Open3D Version Issue) #6

Closed hhcxx2006 closed 11 months ago

hhcxx2006 commented 11 months ago

hi，I am very exciting to your work and want to use your code in our scene. I downloaded Replica datatset and want to test your code first. According to your illustration, I use the command "python3 viz_scripts/final_recon.py configs/replica/splatam.py". The picture below is my final visualization about room0. I find that the reconstruction result seems not correct. This is not the whole room, but only a part of room. What's wrong with this?

1
Nik-V9 commented 11 months ago

Hi, this seems to be a problem with the visualizer. The whole scene's camera trajectory is getting visualized. For some reason, your interactive rendering is not working.

Could you let me know if you're using the latest version of the scripts? Could you share your params.npz so that I could check things on my end?

hhcxx2006 commented 11 months ago

No, the params.npz was generated by your yesterday code. Here is the file, you can check it. https://drive.google.com/file/d/1Py02yvSJZmlnGp3KrTkxRX3tliNtpmh7/view?usp=drive_link

Nik-V9 commented 11 months ago

Thanks for sharing the params. The visualization looks correct on my end.

Replica Interactive Viz

This looks like an Open3D issue on your end. Are you using the correct version (0.16.0)?

Another check would be to add a print statement at this line to check if your code is entering the while loop (interactive rendering):

https://github.com/spla-tam/SplaTAM/blob/b00e6eaa4fd11ea915de4749c422fce7fe10741c/viz_scripts/final_recon.py#L243

hhcxx2006 commented 11 months ago

Hi, thank you for your help. I changed my Open3D from 0.17.0 to 0.16.0. Then it works! Thank you! Now I want to use Intel RealSense D435i to collect my own data. I wonder what inputs should I have? I can collect rgb and depth images. Should I provide camera pose somewhere? How to use my own data? Thank you

Nik-V9 commented 11 months ago

Hi, we have a dataloader for real sense: https://github.com/spla-tam/SplaTAM/blob/main/datasets/gradslam_datasets/realsense.py

You can use this as a starting point to set up the code. The code doesn't require ground truth poses; however, the dataloader requires it. Initializing a list of poses to identity in the dataloader should also work, where the length of the list is the number of frames.

Nik-V9 commented 11 months ago

Closing this issue for now. If you have further queries regarding RealSense, please open another issue.