I get two different values for inf_results when I give input model as ONNX and NEF MODEL

WilliamWilliam
in KL720 related discussion

I wrote inference code for LittleNet model using ktc.kneron_inference . The model is giving me two different inputs for inf_results

For ONNX model ,it is giving me [array([[[[-12.30762 ,  8.2846937]]]])]

For Converted NF model , it is giving me [array([[[[-12.30762 ,  8.2846937]]]])]



Please let me know why this difference in values as both are the same model and if there is an error let me know how to fix it

Tagged:

Comments

  • PowHuangPowHuang

    ONNX is float , Kneron NPU is INT8

    Kneron toolchain make model compression to int8 type (NEF).

    It have some lossy in inference.

The discussion has been closed due to inactivity. To continue with the topic, please feel free to post a new discussion.

© Kneron, Inc.