Qengineering / Face-Recognition-Raspberry-Pi-64-bits

Recognize 2000+ faces on your Raspberry Pi 4 with database auto-fill and anti-spoofing
https://qengineering.eu/deep-learning-examples-on-raspberry-32-64-os.html
BSD 3-Clause "New" or "Revised" License
70 stars 18 forks source link

arcFace ncnn version accuracy and comparing to onnx #7

Open MyraBaba opened 2 years ago

MyraBaba commented 2 years ago

Hi,

Is there any accuracy % for this ncnn recognition model ?

I would like to learn your opinion about the onnx and nccn differences between accuracy (sacrifice if any) and speed for arm & intel cpus ?

ie mobile 0.25 insightface model vs nccn this one ?

Best

Qengineering commented 2 years ago

Dear @MyraBaba,

I haven't done any benchmarking between onnx and ncnn models, nor between ram and CPU, because of the many other projects here in the office.