D3Net: inference on CPU about ai-research-code

[TomonobuTsujikawa]

In previous script, nnabla and openvino were not separated.
So we prepared script that can be run in openvino only environment.
#30

In my environment, original d3net-mss takes 554m49sec by cpu, but openvino (4cpus) takes 8m53sec.

[TomonobuTsujikawa]

Thank you for checking!
Yes, it takes a bit long time. We would like to share how to reduce inference time.
Please give us some time to check and summarize.

[TakuyaNarihira]

We've added an inference script which runs the model on OpenVINO runtime by Intel which should be faster.

See https://github.com/sony/ai-research-code/blob/master/d3net/music-source-separation/README.md#inference-with-openvino or our colab demo to run the OpenVINO model in Python.