doubt about the point cloud in demo video about adop HOT 3 CLOSED

@zuixiaosanlang, I have run this boat dataset through COLMAP, and this is definitely the result of dense reconstruction. This pipeline is not LIDAR-based. It requires cameras, poses, and corresponding points, which you would not be able to get with LIDAR alone.

from adop.

@VisionaryMind
hi, about the scene train, i use default colmap param, i just got 107356 points in sparse reconstruction；
but you got 217407 points.
how to change the colmap param??
thanks

from adop.

@zuixiaosanlang, to get that higher point count, you will need to run preprocess_pointcloud prior to training as follows:

./build/bin/preprocess_pointcloud --scene_path scenes/your_scene_dir --point_factor 2

I have tried 3 and 4, but this creates disturbances in the per-camera neural renders (blurring, color mismatches, etc.). Doubling the point count seems to be the best compromise for sparse camera sets.

from adop.