DensifyPointCloud Failure to reconstruct dense point cloud with ROB-SLAM3 Input

[hplegend]

Call For Help. I use ORB-SLAM3 to explore a Terrasentia-dataset . My purpose is to reconstruct dense point-cloud of Terrasentia-dataset. I follow a instructrument to export KF pose and MapPoints from orb-slam3, and import it to DensifyPointCloud, but it failure to reconstruct dense point clouds. I dispaly the .mvs file and get the following view: I comapred to a standard example and found the "green point" is not in front of camare in my scene just as showing above pic, the programe also throw some exception, the full log as following: """" 21:45:25 [App ] Found a camera not pointing towards the scene center; the scene will be considered unbounded (no ROI) 21:45:25 [App ] Point-cloud composed of 39921 points with:

• visibility info (111936 views - 2.80 views/point): 493 points with 1- views (1.23%) 24538 points with 2 views (61.47%) 7824 points with 3 views (19.60%) 7066 points with 4+ views (17.70%) 1 min / 2.80394 mean (1.8066 std) / 102 max 21:46:22 [App ] Preparing images for dense reconstruction completed: 740 images (56s827ms) 21:46:22 [App ] error: reference image **** has not enough images in view 21:46:25 [App ] Selecting images for dense reconstruction completed: 72 images (3s26ms) 21:46:26 [App ] error: no valid depth-maps found 21:46:26 [App ] Densifying point-cloud completed: 0 points (1m624ms) 21:46:26 [App ] Scene saved (556ms): 740 images (740 calibrated) 0 points, 0 vertices, 0 faces 21:46:27 [App ] MEMORYINFO: { 21:46:27 [App ] VmPeak: 1623344 kB 21:46:27 [App ] VmSize: 467292 kB 21:46:27 [App ] } ENDINFO """"

[cdcseacave]

is the camera and 3d points positions in Viewer what you would expect? I suspect the way you export the SfM data to MVS is not correct

[ddkats]

@hplegend What type of input data did you parse to the DensifyPointCloud?