BeagleY AI- Low FPS and Delegate Identification

Dear Experts,

We are trying to use the BeagleY-AI board for object detection purposes. For the initial test, we followed the same procedure as given in the demo using TensorFlow lite. The output was obtained using a USB camera at the fps rate of 3. We would like to increase the performance as the theoretical capabilities is higher and up to 4 TOPS.

Any suggestions or demo would be appreciated regarding the delegate files for hardware acceleration or any specific SDK files to be installed.

That tutorial needs to be updated – it’s running CPU only as Beagleboard Debian does not have hardware acceleration.

You my have better luck starting out with the TI SDK Edge AI image we just ported https://www.beagleboard.org/distros/beagley-ai-ti-sdk-edge-ai-11-00-00-08-2025-07-03

It runs the sample apps using c7x hardware acceleration and ISP support. I’d like to update the tutorials to work off TI Edge AI image at some point.