CPU Utilization is high, even if the model is running in GPU architecture.

[tarunsk1998]

Hi,

I have a Nvidia GPU(Tesla V100), on which I was trying to run the Detecto(model_name="fasterrcnn_resnet50_fpn").
I noticed that the model is able to use the GPU, but I am seeing a huge hike in the CPU utilization as well.
Is this expected behavior? If so why is this happening and how can I reduce the overload on the CPU?

GPU memory - 1934 MB
GPU Utilization - 33% (average)
CPU Memory - 2250 MB
CPU Utilization - 75 - 99% -------> This is what I am concerned about.

This is how I initialize the Detecto model,
from detecto.core import Model
device = torch.device('cuda') if torch.device.is_available() else torch.device('cpu')
detecto_model = Model(model_name="fasterrcnn_resnet50_fpn", device=device)

Other useful information,
detecto version - 1.2.2
torch version - 2.0.1+cu117