search

lppllppl920 / EndoscopyDepthEstimation-Pytorch

Official Repo for the paper "Dense Depth Estimation in Monocular Endoscopy with Self-supervised Learning Methods" (TMI)
GNU General Public License v3.0
127 stars 28 forks source link

Try a new data set #25

Closed caozhengtao110 closed 4 years ago

caozhengtao110 commented 4 years ago

Dear Author： Follow the requirements in your readme.md. If you change to a new data set, I need to preprocess the data set. I need to get motion.yaml through SFM or SLAM. In addition, there are camera_intrinsics_per_view, selected_indexes,visible_view_indexes Do I need to get structure.ply before I train? Does it represent the reconstruction result of one frame or the reconstruction result of multiple frames?

lppllppl920 commented 4 years ago

Hi Zhengtao,

Yes, you need to get the data like the format here yourself. structure.ply is generated from SFM or SLAM.

YanhaoZhang commented 4 years ago

Hi @lppllppl920, I think it might be good to mention that the pose definition from colmap is from world to camera, while in your code, it is from camera to world. Just in case someone else is going to use colmap to preprocess the image.

caozhengtao110 commented 4 years ago

Thank you for the author and Thank you for friend

qqc222111 commented 9 months ago

Hi @lppllppl920, I think it might be good to mention that the pose definition from colmap is from world to camera, while in your code, it is from camera to world. Just in case someone else is going to use colmap to preprocess the image.

So what should I do if I use colmap to preprocess the data in order to generate correct depth map