I’m thinking if I could add gaze estimation to my existing c++ project. Seems that if I would be running with Python the choose could be GazeML or iTracker? But since I’m doing this with C++ I was thinking if I could use YoloV5 for this? And if so, should I download model “yolov5s.pt” and convert it to ONNX? Or am I totally lost here?
how would you use an object detection model for gaze estimation ?
it does not make any sense
Ok, good point. Understood something wrong when I was scrolling related samples on the web. Sorry.
Any hints what could work here?
well, the ‘old school’ idea, off my head would be something in the line of:
- face detection
- landmarks
- 3d headpose (solvePnP)
- physical ‘raycasting’ through the pupils
- then combine tranformations
however, i’m almost sure, nowadays you can find a cnn, doing all of this in one go …