Hello, I have a project where I want to use the ofxTensorFlow2 for pose detection. But it has to run on relatively old apple computers, so no chance to use any cuda-gpu-based approach.
I’ve found that the tfjs has a webgl backend, and runs reasonably fast even on old devices.(over 50fps)
I have built a simple web application and tried to grab the results(segmentation and skeleton) from OF.
I could possibly grab it by using a virtual webcam, syphon and the skeleton data could be sent via OSC.
But after running them all together, my OF app was slowing down, so I was wondering if there is a better approach to use pre-trained models on devices without cuda gpus.
I wonder if anyone of you has a similar experience or a better approach.