How well does Jetson Orin Nano handle facial recognition and anti-spoofing?

[kxnglish]

Hi everyone,

I’m planning to build a facial detection system using the NVIDIA Jetson Orin Nano and would like to understand its capabilities in this area. My project will involve both facial recognition and anti-spoofing measures (e.g., liveness detection).

I have a few questions:

1. How effective is the Orin Nano for real-time facial recognition? Are there any recommended models (e.g., FaceNet, ArcFace, InsightFace) that work well with the Orin Nano for high accuracy and speed?
2. Is the Orin Nano powerful enough for anti-spoofing techniques like texture analysis, depth detection (using additional sensors), or blink/head movement detection? Has anyone tried implementing these on the Orin Nano, and what were your results?
3. What frameworks or tools do you recommend for optimizing models for better performance on the Orin Nano (e.g., TensorRT, DeepStream)?
4. Any specific tips for handling video streams from IP cameras, such as Ubiquiti AI Cameras, for smoother integration with DeepStream?

I’m looking for insights based on your experiences—any suggestions, model recommendations, or tips on improving accuracy and response times would be greatly appreciated!

Thanks in advance!

[Fiona.Chen]

@kxnglish

• How effective is the Orin Nano for real-time facial recognition? Are there any recommended models (e.g., FaceNet, ArcFace, InsightFace) that work well with the Orin Nano for high accuracy and speed?

Nvidia does not provide any face recognition models according to the law. There is no recommendation for the model.
There are just some pre-trained face detection models provided by TAO toolkit. Overview - NVIDIA Docs

@kxnglish

• Is the Orin Nano powerful enough for anti-spoofing techniques like texture analysis, depth detection (using additional sensors), or blink/head movement detection? Has anyone tried implementing these on the Orin Nano, and what were your results?

It depends on your algorithm and how did you accelerated your algorithm with GPU.

@kxnglish

• What frameworks or tools do you recommend for optimizing models for better performance on the Orin Nano (e.g., TensorRT, DeepStream)?

TensorRT is the model inferencing acceleration tool. DeepStream is a video/audio inferencing framework SDK. DeepStream also uses TenorRT to accelerated the model inferencing task.

@kxnglish

• Any specific tips for handling video streams from IP cameras, such as Ubiquiti AI Cameras, for smoother integration with DeepStream?

DeepStream supports most standard live streams such as RTSP/HTTP,…

If your Ubiquiti AI Cameras provides the standard RTSP(RFC 2326: Real Time Streaming Protocol (RTSP)) streams, DeepStream can support it.

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks