NPU Compatible Facial Landmarks Model

I would like to try out facial landmarks detection on the NPU.
In parallel, I am trying to convert existing TensorFlow models to the NPU format but with no success till now.
Is there an available NPU compatible model that provides 68 landmarks that I can use?

Hi, are you using Khadas branded camcorder?

No, I’m not using OS08A10 8MP HDR Camera.
Why is it important?

Does your camera have an Interface: MIPI-CSI?

The more information you can give about your equipment, the more likely you are to get a complete answer.:slightly_smiling_face:

@Vladimir.v.v its the same camera, that’s not the question at matter here,

@NoamZ, the khadas team can answer your question, but they are not here for a few weeks as they are on spring vacation,

ping @Frank for more info once they come back,

also please mention the tensorflow version, any error logs if possible, some other experienced NPU user can reply back to you by then,


No relevant logs. I’m just trying to find facial landmarks model that is compatible with the NPU.
Looking forward to hear from the Khadas team.

1 Like

By the way, while the Khadas team has a weekend, you can browse the Khadas forum and use the forum search!
good luck to you!

I’m currently testing a model for the VIM3 NPU (face_landmark_68_88.nb).

To get the model do:

git clone GitHub - khadas/AML_NN_SDK && cd AML_NN_SDK

$ find . -name ‘_68

Note: the model for VIM3L should be face_landmark_68_99.nb

1 Like


VIM3 : AML_NN_SDK/face_landmark_68_88.nb at master · Amlogic-NN/AML_NN_SDK · GitHub

VIM3L: AML_NN_SDK/face_landmark_68_99.nb at master · Amlogic-NN/AML_NN_SDK · GitHub