I used a pretrained ENET-like model in .pb format for image segmentation on traffic, and it used up 1.8 GB of my RAM.
I'm aware that deep-learning models take up a lot of VRAM space, but this is the first time i have seen a model using this many memory on my CPU. I tried to limit the memory on GPU ,but it didn't have any effect on RAM usage. Has anyone come across this issue?
Specs: 16GB RAM, Geforce GTX 1050 with 2GB VRAM, tensorflow 2.3.1 Also i ran the inference on compat mode.