search

ingra14m / Deformable-3D-Gaussians

[CVPR 2024] Official implementation of "Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction"
https://ingra14m.github.io/Deformable-Gaussians/
MIT License
884 stars 49 forks source link

dataset #25

Open Tiandishihua opened 9 months ago

Tiandishihua commented 9 months ago

How do I make my own video dataset?

ingra14m commented 9 months ago

Hi, if it's a multi-view dynamic video, I recommend using IBR methods, such as im4d, 4k4d. If it's a single-view dynamic video, the essence of our method is to obtain the frame and its corresponding camera pose.

A direct approach is to use colmap to force a match, but this may carry some potential risks leading to inaccurate camera poses (such as with the HyperNeRF dataset). I also recommend Record3D for estimating the pose of dynamic scenes.

Lwy-1998 commented 9 months ago

Hi, if it's a multi-view dynamic video, I recommend using IBR methods, such as im4d, 4k4d. If it's a single-view dynamic video, the essence of our method is to obtain the frame and its corresponding camera pose.

A direct approach is to use colmap to force a match, but this may carry some potential risks leading to inaccurate camera poses (such as with the HyperNeRF dataset). I also recommend Record3D for estimating the pose of dynamic scenes.

Sorry, could you please provide a link for Record3D?