VRAM Consumption?

[CultureAddiction]

In paper, It is written that VRAM consumption is way lower than others, 256mb or somethin.
But when I train bicycle dataset with command below, I checked nvidia-smi and it was consuming vram about 1~2GB, and its getting more and more when it goes to next step. It was over 7 GB at the end. Is that other way to check VRAM consumption? or just me having such issue?

~/YBS/torch_merf/torch-merf$ python main.py /data_ssd/YBS/merf_data/bicycle/ --workspace trial_bicycle --enable_cam_center --downscale 4