search

Qengineering / Face-Recognition-with-Mask-Jetson-Nano

Recognize 2000+ faces on your Jetson Nano with additional mask detection, auto-fill and anti-spoofing
https://qengineering.eu/deep-learning-examples-on-raspberry-32-64-os.html
BSD 3-Clause "New" or "Revised" License
35 stars 6 forks source link

How to run with GPU #7

Closed lbtanh closed 3 years ago

lbtanh commented 3 years ago

Thanks for great repo, I wonder if we can run with GPU or both? Thanks again

Qengineering commented 3 years ago

Dear Ibtanh,

Thanks for your compliments. You are using two frameworks here. Paddle-Lite which already uses CUDA (=GPU) ncnn which uses not CUDA but Vulkan. As you know, Vulkan is also GPU based software used to accelerate matrix operations.

lbtanh commented 3 years ago

Dear Ibtanh,

Thanks for your compliments. You are using two frameworks here. Paddle-Lite which already uses CUDA (=GPU) ncnn which uses not CUDA but Vulkan. As you know, Vulkan is also GPU based software used to accelerate matrix operations.

Dear @Qengineering,

Thanks for quick reply, I have run the program successfully, but i dont see any GPU was used (monitor with jtop command). i'm just curious at that

Qengineering commented 3 years ago

When I run the app, jtop shows GPU usage, on and off. Up to 50%. In the main loop there's a lot of code not using the GPU, so no surprise when jtop samples the GPU it doesn't see a continue flux. However, running the movies, it must show something. Otherwise it may indicate a installation of ncnn / Paddle-Lite without Vulkan/CUDA on your Nano.