CUDA GPU memory continuously increases when running superpoint algorithm

[erdemkkok]

CUDA GPU memory continuously increases when running superpoint algorithm ?

I have tried cuda.empty.cache method already. How can i fix that issue ?
i think that issue about the torch library