How to train BeagleBone AI to Recognize Humans

I followed this tutorial for running a classification program on the BeagleBone AI:

I was very pleased with the results. When I looked through clasification.tidl.cpp, I noticed this block of code:

selected_items[0] = 429; /* baseball /
selected_items[1] = 837; /
sunglasses /
selected_items[2] = 504; /
coffee_mug /
selected_items[3] = 441; /
beer_glass /
selected_items[4] = 898; /
water_bottle /
selected_items[5] = 931; /
bagel /
selected_items[6] = 531; /
digital_watch /
selected_items[7] = 487; /
cellular_telephone /
selected_items[8] = 722; /
ping-pong_ball /
selected_items[9] = 720; /
pill_bottle */

I assumed this is what determines which items from a list the program will try to identify. The program also references a file called ‘imagenet.txt’, which I assumed was the list of every object the neural network was trained to recognize. I looked through imagenet.txt, but I couldn’t find any ‘person’ or ‘human’ objects. This means I will either need to train the beaglebone myself or somehow find a new model. How can I make it so that the example program can identify people?

The BeagleBone AI is only performing “inference” using an already trained model. No training or learning is occurring on the BBAI itself.

You may want to start with the introduction video linked from the tutorial at:

To train a new model you will need to learn the Caffe or TensorFlow frameworks. This is unrelated to the BeagleBone hardware itself.

You can find more technical information about what types of models are supported by the TI API here:

Thanks! I will look into those frameworks.

Thanks for the clarification! Which framework does the sample model use?