Giters

ExplainableML / sketch-primitives

ECCV 2022: Abstracting Sketches through Simple Primitives

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

PMN model took up a large computing resource

WinKawaks opened this issue · comments

commented

When I tried to process the dataset with PMN to create a new dataset consisting of primitive reconstructions with the following command, I got a CUDA OOM error.

python save_processed_images.py --log-name pmn_quickdraw09_processed --model-type pmn --dataset quickdraw09 --test ./log/pmn_quickdraw09/model_best_acc.pt --only-save-coords

Even I set the batch-size=1, I still got the error. My GPU is 2070. I want to know why generating new samples with PMN is more resource intensive than training PMN. Could you please give me some advice on how to reduce the CUDA memory?