search

Matrix-ASC / DeLA

Official PyTorch implementation of DeLA
MIT License
41 stars 5 forks source link

The hash_size in grid_subsampling #4

Closed HaoRan-hash closed 11 months ago

HaoRan-hash commented 11 months ago
屏幕截图 2023-10-20 161300 屏幕截图 2023-10-20 161313

I would like to know how downsampling ratio was determined and how it relates to grid size.

binjiechen commented 11 months ago

Hi, grid subsampling picks one point in each grid. A larger grid size leads to smaller grids and thus a higher downsampling ratio (original / downsampled points). The actual downsampling ratio depends on the data. I simply load some point clouds, grid sample (leaving hash_size to 1) them, and print the ratios, then determine one. This parameter setting is mainly for performance issues. It's been quite some time since I wrote it and I can't recall its impact. But it should be minor. Just don't set it too small, in which case the hash table can't accommodate all subsampled points and a runtime error can occur.