search

med-air / EndoNeRF

Neural Rendering for Stereo 3D Reconstruction of Deformable Tissues in Robotic Surgery
https://med-air.github.io/EndoNeRF/
173 stars 14 forks source link

cannot generate pose_bounds.npy for cutting_tissues_twice #2

Open peiran88 opened 1 year ago

peiran88 commented 1 year ago

When I tried to use the images folder in cutting_tissues_twice dataset to generate pose_bounds.npy by myself, I met this error:

image

But I can generate correct pose_bounds.npy for testscene dataset followed the instructions of LLFF using docker. Does that mean there are some problems in the dataset so It cannot generate poses? How did you generate pose_bounds.npy file? @yuehaowang

yuehaowang commented 1 year ago

Hi Peiran, I've already provided a 'pose_bounds.npy' file in the open dataset. No need to generate camera poses.

peiran88 commented 1 year ago

Hi Peiran, I've already provided a 'pose_bounds.npy' file in the open dataset. No need to generate camera poses.

I can successfully train these two datasets and generate point clouds today using your provided pose_bounds.npy file. I just have a try to generate pose_bounds.npy and it may help me to test other datasets. I am confused about this error. Did you ever meet this problem and did you make some changes to generate this pose_bounds.npy file?

yuehaowang commented 1 year ago

Hi Peiran, I've already provided a 'pose_bounds.npy' file in the open dataset. No need to generate camera poses.

I can successfully train these two datasets and generate point clouds today using your provided pose_bounds.npy file. I just have a try to generate pose_bounds.npy and it may help me to test other datasets. I am confused about this error. Did you ever meet this problem and did you make some changes to generate this pose_bounds.npy file?

I guess you are using the 'imgs2poses.py' script from LLFF to generate camera poses. Since our setting is single-viewpoint, I didn't use this script to generate camera poses but just simply set camera poses to identity matrices (no rotation & no translation) and then save them in LLFF data format.

Regarding the problem you met, I think you can first remove the existing 'pose_bounds.npy' file and then run the script again. Nevertheless, my previous observation tells colmap may not work very well on our data and could result in very bad performance :(

peiran88 commented 1 year ago

Thank you so much for telling me that😊