Face_recognition npu example fails to run

When trying to use the face_recognition npu example, the retinaface network file fails to load

sudo ./face_recognition -M …/data/model/retinaface_int8.adla -m …/data/model/facenet_int8.adla -p 1
invalid/unsupported operator: 45

I have successfully run other examples, such as yolov7_tiny_cap on the host

I also get the same error for the facenet example

sudo ./facenet -p 1 -m …/data/model/facenet_int8.adla
invalid/unsupported operator: 45
E NN_SDK:[aml_adla_create_network_common:357]Error: create network fail.
amlnn_init is fail
init_network fail.

Hello @davidharding,

@Louis-Cheng-Liu will help you here.

Hello @davidharding ,

I am sorry that Demos 3,4,5,6,7 are not suitable for Linux kernel 5.4 version. They will be supported in next version, 5.15.

Do you have a VIM4 image based upon Linux Kernel 5.15 available for download, there doesn’t appear to be one on either dl.khadas.com - Index of /.images/vim4/ or dl.khadas.com - Index of /products/vim4/firmware/
Or if I need to build one myself using GitHub - khadas/fenix: One-stop script set to build Ubuntu/Debian images is there a proven working debian distribution I can base it upon?

We will release the 5.15 kernel image for VIM4 next week, you can check at that time.

Can I be cheeky and ask if it will be uploaded to Fenix - on a Debian base? :smiley:

Is the 5.15 kernel still due to be shared this week, or do you have a revised timeline?

Hi @davidharding @RichardPar, Please check the release

https://forum.khadas.com/t/vim4-4n-ubuntu-22-04-linux-5-15-testing-images-are-available-now/20447/2

1 Like

Awesome… Thank you!

are there plans to get it in to the Fenix Github ? I would like to do a Debian build.

Richard

Not yet at the moment, this is a preview image just to test the kernel and iron out any bugs and is only a pre-release.

You can expect the sources to arrive for your own builds maybe a couple of weeks.

Regards.

The NPU examples don’t appear to run on the ubuntu images provided
We now get the following error for yolov7_tiny, which worked on the previous kernel
The face recognition also gives the same error

sudo ./yolov7_tiny -p …/data/horses.jpg -m …/data/yolov7_tiny_int8.adla
adla usr space 1.2.0.5
E NN_SDK:[aml_adla_create_network_common:357]Error: create network fail.
amlnn_init is fail
init_network fail.
root@Khadas:~/khadas/v

@davidharding developers are currently on vacation, they will get back to you on this issue as soon as they return, kindly requesting your patience.

Regards.

Hopefully refreshed from Golden Week and ready to fix bugs :smiley:

1 Like

Hello @davidharding ,

Which image do you use? I use 5.15 kernel image and all examples can run normally.

I was already using the ubuntu desktop kernel 5.15 image, but I have rebuilt my system again this morning, and I have 4 of the 5 “face” demos working now.

I would like to do a Debian build of 5.15 kernel now using Fenix, rather than Ubuntu.
Would it be possible to push the test image changes to a branch in github?

Hello @davidharding

The source code is not ready at this moment.

Is there any timeline for a beta of the code? I dont mind helping :smiley:

Any progress with the code?