Best way to calculate the Memory Required for OpenVINO model Inference?

[anujdutt9]

Hi Everyone. I have a model converted from PyTorch to ONNX to OpenVINO. I am able to run the model on NCS2 using OpenVINO as well.
The question I have is, is there a good way to estimate the DRAM bandwidth required for the deployed model for inference on NCS2?
I know the model input size and I know the model parameters size i.e. model size in MB.
The estimated memory required at runtime should be a sum of the input image size, model weights and activations as input passes through each layer till the end.
Can someone please help/guide me as to how to get this estimate for a OpenVINO model on NCS2 ?
Thanks.

[brmarkus]

For a first estimate you might want to check with the Model-Analyzer "https://github.com/openvinotoolkit/model_analyzer".