NPU SDK: How to build the libnn_xxxx.so library for custom model converted with Amlogic converter

VIM3
1

Hi! I have VIM3 and I would like to benchmark my custom model on the NPU.

I follow the documentation guide on Model Conversion.

I have already quantized and converted the model using acuity-toolkit and got nbg_unify_xxxx directory with files. But I don’t know how to build the library libnn_xxxx.so from these files to run inference using KSNN API. Is there any way to build it other than using the KSNN Convert tool?

Thank you!

2

@inarm You can found the demo source in this page

Then, you can convert follow this doc

Using ksnn is very simple. As long as you get data with a few simple APIs, you can compare the source code of different demos. There is only the difference between pre-processing and post-processing. The conversion and use of the api are exactly the same.

3

Thanks, I used the conversion tool and reconverted the model.
But is there a way to add data for quantization? As far as I see there is no parameter —source in convert script to specify data on which the quantization should be done

4

@inarm aml_npu_sdk/acuity-toolkit/python/data/dataset

If you are a single-input model, fill in the data in dataset0.txt