Calculating memory limits and adjusting images to fit in memory

[felixgeen]

Uncaught exception: Could not allocate memory: E:\instant-ngp\dependencies\tiny-cuda-nn\include\tiny-cuda-nn/gpu_memory.h:112 cudaMalloc(&rawptr, n_bytes+DEBUG_GUARD_SIZE*2) failed with error out of memory

Am I correct in thinking that the contributing factors of having a dataset fail due to a memory limitation are based on these two variables:

1. the resolution (dimensions) of the individual images
2. the quantity of images in the data set

Are there any other factors that contribute to the chances of failing to launch the testbed? is there any way of calculating ahead of time if the dataset will fail? For example, can I add up the resolutions of the dataset and workout if it will fit into memory? How would I do this?

I have a 3080ti with 12 GB of VRAM.

Alternatively, if it is not possible to calculate ahead of time: Is there any way of resizing the images in the dataset without having to re-compute the COLMAP again? Because when I try downscaling my images, the testbed loads, but it looks totally wrong because presumably there are some other details inside the transforms.json that contribute to resolution besides the "w" and "h"?