If I run custom data, how many RGBs and depths do I need?

NVlabs / BundleSDF

[CVPR 2023] BundleSDF: Neural 6-DoF Tracking and 3D Reconstruction of Unknown Objects

https://bundlesdf.github.io/

Other

1.07k stars 118 forks source link

If I run custom data, how many RGBs and depths do I need? #150

Closed GavinYang5 closed 6 months ago

GavinYang5 commented 6 months ago

Thank you for your excellent work! I want to try getting Mesh on my own dataset. How many RGB and depth images do I need, and if I have a mask for each image, do I still need to ensure that the positions of objects in the image are continuous over time, similar to video streams? Also, if I use a higher precision RGB camera, the effect should be better, right?

wenbowen123 commented 6 months ago

Hi, this work assumes there is a continuous RGBD stream. It only needs the images and will auto compute the poses. If you have discrete images and you already have the poses for them, you can try the model-free setup in FoundationPose.