search

CUHK-AIM-Group / EndoGaussian

EndoGaussian: Real-time Gaussian Splatting for Dynamic Endoscopic Scene Reconstruction
https://yifliu3.github.io/EndoGaussian/
MIT License
109 stars 5 forks source link

COLMAP method #27

Open Beilinlin opened 2 months ago

Beilinlin commented 2 months ago

Hello! I noticed in your paper that the point cloud initialization methods include Random, COLMAP, and the back-projection method you used. I would like to ask how I can use the COLMAP method to initialize the point cloud. Could you provide a general overview of the steps involved? Thank you very much!

yifliu3 commented 1 month ago

Hi! I have tried using the COLMAP for initialization according the repo: https://github.com/hustvl/4DGaussians, where they use colmap in nerfstudio for quick implementation. After getting corresponding files, you can directly load the data according to their colmap dataloader.

Beilinlin commented 1 month ago

Thanks very much! I used three images from the cutting and scared datasets and was able to generate a sparse point cloud using COLMAP, but when I used three images from the pulling dataset, the point cloud generation failed. Have you encountered this problem before?