search

facebookresearch / consistent_depth

We estimate dense, flicker-free, geometrically consistent depth from monocular video, for example hand-held cell phone video.
MIT License
1.61k stars 236 forks source link

Using ORB_SLAM for camera pose computation #38

Open lkosh opened 3 years ago

lkosh commented 3 years ago

Hi, I've been trying to pre-compute camera poses of TUM_RGBD dataset with ORB_SLAM2 tool and running consistent_depth algorithm. Ideally, this should result in a faster and more accurate reconstruction, since SLAM is better at predicting poses. However I haven't been able to get consistent_depth to work with these poses, because COLMAP fails to reconstruct depth maps.

In the paper you mention that you have tested the algorithm on TUM_RGBD dataset, which is close to what I'm trying to do, since ORB_SLAM outputs poses in TUM dataset format. Could you share the steps you took to achieve a successful evaluation on this dataset?

Robertwyq commented 3 years ago

I have the same question

Robertwyq commented 3 years ago

Hi, I've been trying to pre-compute camera poses of TUM_RGBD dataset with ORB_SLAM2 tool and running consistent_depth algorithm. Ideally, this should result in a faster and more accurate reconstruction, since SLAM is better at predicting poses. However I haven't been able to get consistent_depth to work with these poses, because COLMAP fails to reconstruct depth maps.

In the paper you mention that you have tested the algorithm on TUM_RGBD dataset, which is close to what I'm trying to do, since ORB_SLAM outputs poses in TUM dataset format. Could you share the steps you took to achieve a successful evaluation on this dataset?

Did you solve the problem?