Hi all,
I’m working on a face recognition attendance project using Luckfox Pico Max. I’m using the Luckfox team’s example that runs RetinaFace + MobileNet models simultaneously. For the user interface, I plan to integrate a 2.8" ResTouch LCD with touch input.
My main concerns are:
1. How well does the 2.8" ResTouch touch screen perform on the Pico Max when these models run together?
2. Is there any noticeable lag or performance drop in either the AI processing or the touchscreen responsiveness?
If anyone has experience with this setup or similar configurations on the Luckfox Pico Max, I would really appreciate any insights, tips, or recommendations.
Thanks in advance for your help!
Performance Impact of Running RetinaFace + MobileNet with 2.8" ResTouch LCD on Luckfox Pico Max?
Hello,
1 The demo we provide displays inference results on the 2.8" ResTouch screen at an efficiency of approximately 10 fps. Using multithreading to optimize the demo could lead to better performance.
2 AI inference not only utilizes the NPU but also relies on the CPU for some graphical processing. Since the RV1106 has only a single-core CPU, increased CPU usage will inevitably lead to higher system latency. Touch input itself does not consume much CPU scheduling; the main factor is the processing function of the touch action, which should be analyzed based on the actual situation.
1 The demo we provide displays inference results on the 2.8" ResTouch screen at an efficiency of approximately 10 fps. Using multithreading to optimize the demo could lead to better performance.
2 AI inference not only utilizes the NPU but also relies on the CPU for some graphical processing. Since the RV1106 has only a single-core CPU, increased CPU usage will inevitably lead to higher system latency. Touch input itself does not consume much CPU scheduling; the main factor is the processing function of the touch action, which should be analyzed based on the actual situation.
Thank you for reply.Crocodile wrote: ↑2025-05-16 10:30 Hello,
1 The demo we provide displays inference results on the 2.8" ResTouch screen at an efficiency of approximately 10 fps. Using multithreading to optimize the demo could lead to better performance.
2 AI inference not only utilizes the NPU but also relies on the CPU for some graphical processing. Since the RV1106 has only a single-core CPU, increased CPU usage will inevitably lead to higher system latency. Touch input itself does not consume much CPU scheduling; the main factor is the processing function of the touch action, which should be analyzed based on the actual situation.
Anyway,
We are using the RV1106 with RetinaFace + MobileNet for face detection. To improve security, we want to add liveness detection (e.g., using IR or thermal cameras).
Does RV1106 support liveness detection?
Do you have any recommendations or examples (e.g., with dual-camera setup: RGB + IR)?
Thank you!
Best regards,
We haven't tested the live detection solution yet, so we can't determine its feasibility. You need to adapt the infrared thermal imaging camera and the convolutional neural network model suitable for infrared images. Both of these are not easy to implement.