search

TUTvision / MuSHRoom

Indoor room dataset used for novel view synthesis and 3d reconstruction (WACV 2024)
34 stars 1 forks source link

Processing raw kinect data #6

Open Haven-Lau opened 4 weeks ago

Haven-Lau commented 4 weeks ago

Hi there, Thanks for the awesome dataset!

image

I was going through the paper and saw the 3.1.2 section about post processing the raw kinect dataset, is everything described here included in the kinect tools folder? I'm trying to learn the steps to reproduce the dataset output from raw kinect videos, any help would be appreciated.

Thanks again

XuqianRen commented 4 weeks ago

Hi @Haven-Lau , thanks for your interest in this dataset. The kinect tool folder includes 1. extract keyframes from the raw kinect video, 2. split train and test set, 3. get point cloud only from train set/ whole set. 4. render depth from point cloud get from 3, 5. complete kinect depth with the depth value got from 4.

I didn't put the scripts to get the colmap pose from the kinect sequence, and align the colmap pose with the spectacular AI pose. This step is used to register the long and short sequence, because I found directly register long and short sequence's spectacularAI pose without refinng each keyframe pose will lead to very bad alignment, so I register all images again with colmap. If you want to know how to do this alignment I can also share my script.