Memory fills up (CPU OOM -> killed)

[thegenerativegeneration]

Hi, I wanted to try out the reconstruction, but for me the method does not go beyond the data loading of the colmap dataset.

The concatenation of the arrays leads to CPU memory filling up which ends up in a crash.

I have 64GB of memory.

I think the way the data is loaded is problematic as all data of the dataset is loaded into memory, as far I understood the code.

The crash happens exactly at

Instant-angelo/datasets/colmap.py

Line 307 in bb7383a

self.all_fg_indexs = torch.cat(self.all_fg_indexs, dim=0)