SlowFast Inference: GPU CUDA out of memory

Hello,

I am using SlowFast model for inference on a GPU machine with 16 GB memory. When I load the model it is taking 1.5GB GPU memory. While performing inference it is gradually increasing and throwing CUDA out of memory error. I believe GPU is not releasing the memory and not issue with SlowFast model.

Can someone explain how to resolve this? If there is some other issue for this, please let me know the root cause and how to fix this.

Thanks in advance.