nef file run in KL630 flash

YOUNGJUN YOOYOUNGJUN YOO
in KL630 related discussion

I converted 3DCNN - video classificatin model to nef as follows:

However, when I run this nef file in mnt/flash/vienna/nef/  and change the modelPath in

mnt/flash/vienna/ini/host_stream.ini  KL630 board. I got following errors:



How can I handle this problem?

I modified the host_stream.ini from yolo model but my custom model has no ref. So, I think this file might cause the problem. I also atteched the host_stream.ini file, too.


Please help me.


 

Comments

  • Nick WangNick Wang

    Hi YOUNGJUN,

    Has your model been simulated in the toolchain docker?

    It is recommended to follow the workflow on the website and confirm that there is no problem with the model before porting to KL630.

    https://doc.kneron.com/docs/#toolchain/manual_1_overview/

  • YOUNGJUN YOOYOUNGJUN YOO

    Thanks a lot Nick for answer.


    I already simulated tool chain docker with kl630 with following code


    which Corresponds to 1.4~1.6.2. Hardware Simulation step.

    The result of the above code is

    image.png


    Above results shows binary_inf_results is almost same as fixed_point_inf_results.

    Please give me advice on what I can do next.

  • Maria ChenMaria Chen

    Hi Youngjun,

    Your binary_inf_results and fixed_point_inf_results should be almost the same, so it seems successful.

    image.png

    According to your screenshot, it seems like your nef file is generated. It depends on where you set the nef path, but it is usually in /data1 folder.

  • YOUNGJUN YOOYOUNGJUN YOO

    Dear Maria Chen,


    I found the weight file, and

    Thank you for your kind reply.

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

© Kneron, Inc.