search

CUHK-AIM-Group / EndoGaussian

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

Camera Pose Optimization #10

Closed Gaozhongpai closed 6 months ago

Gaozhongpai commented 6 months ago

Thanks for opening source this work. May I ask if the current method can be used for endo videos with camera movements? During the training, does this method optimize the camera pose as well? Since if the initial poses from COLMAP may not be very accurate, the overall performance is not good. I notice that the EndoNeRF dataset's camera doesn't move but some other endo videos may have camera motions. Thanks again.

yifliu3 commented 6 months ago

Thanks for your attention. The current method can be used with camera movements, and we have tested on the SCARED dataset, which consits of static scenes with moving cameras.

For the setting of the surgical reconstruction, we follow previous work to use the gt camera pose, thus the pose is not optimized like some SLAM of SfM methods. As the gt camera pose is provided by the dataset, we don't need to use COLMAP to estimate the pose. And the initial points are generated by a direct re-projection (More details can refer to the paper).