Hello @longer

You can use yolo-face model instead of faceNet. However, this model only detects face box without key pionts and only has python without C++.
If you want to use it, please refer to KSNN Usage [Khadas Docs].
If not, please try to convert your model. Refer to NPU SDK Usage [Khadas Docs].

@Louis-Cheng-Liu please help me!!

why not working…

Hello @rayeleigh ,

Are you sure the name of your model is converted_model.tflite.tflite but not converted_model.tflite.



I tried changing it but it doesn’t work
I don’t think my filename is wrong

Hello, @longer ,

Yes, yolo-face only can detect face.
Sry, I do not participate in the face recogniton project. I can suggest nothing.

Hello @rayeleigh ,

Could you show your

#! /bin/sh


if [ ! -e “$pegasus” ]; then

$pegasus import tflite
–model converted_model.tflite
–output-model converted_model.json

Hello, @rayeleigh ,

Is the code is copied from the I copy your code and find some symbols are not English. It can run after i modify them. Here is the code i modify and the result i run.


if [ ! -e "$pegasus" ]; then

$pegasus import tflite \
    --model converted_model.tflite \
    --output-model converted_model.json \

@Louis-Cheng-Liu uis-Cheng-Liu
I copied and entered the code but it doesn’t work…

I ran 0._bash.import right after downloading the sdk tool, is it wrong??
Give me your email and I’ll send you my files.
my email is
can you help me?

Hello @rayeleigh ,

From your description, i think you forget to install the tool. Please try again after installing.

This picture is from aml_npu_sdk/docs/en/Model Transcoding and Running User Guide (1.0).pdf. Some more converting details can be found in it.

If still fail, please send your files for me. My e-mail is

hi i read it
but i have problem…
what is it?

i don’t have reqirement.txt…
please help me

Hello @rayeleigh ,

Create a requirements.txt and write in as follows.

Please use tensorflow==2.0.0a0 instead of tensorflow==2.0.0.

Have you remembered how you download the tool? Git clone or download zip in windows?

yes!!! i remember
i use git clone
I need to create a txt file

I need your advice!

I’m making a Korean sign language translator in Python language

The gui is using the tkinter program
The real-time recognition program is using open cv.

Is it not the right choice for me to use the c/c++ sdk now?

Should I use open cv dnn?
By the way, what is the difference between python ksnn and opencv dnn??

Also, what are the advantages of using opencv dnn??

Once these programs are used, npu can be used, but I have to make an inference process, right?
@Louis-Cheng-Liu @numbqq @goenjoy

Hello @rayeleigh ,

I advise you to use C++. C++ can interface npu directly.
Both opencv dnn and ksnn provide python API. But the API is different.
If you want to use ksnn, you should convert model first. But opencv dnn can use onnx model.(Onnx model still need to be quantized)
Opencv dnn is more general. Ksnn is easier but only be use in Khadas.
Yes, after you convert model, you should make an inference process in C++.

@Louis-Cheng-Liu Do you know the difference between the versions??

raw or not raw
Speaking of dates, these are two files from April 25
i i know gnome and server

Hello @rayeleigh ,

Versions with raw is installing by oowow. Without raw is installing by usb.

I suggest you to install by usb.

@Louis-Cheng-Liu good
and Open cv dnn should also be done on a computer?

I’m almost on my way to the answer. thank you for helping me