Step - 0: Creating a base for the model by experimenting with mediapipe

1st Problem:

MediaPipe loads its own DLL (libmediapipe.dll) using Python's ctypes It then tries to find a C function called free() inside that DLL On Linux/macOS, free() is available everywhere, it's part of the standard C library bundled into shared libraries On Windows, free() is not inside libmediapipe.dll. It lives in a separate system DLL called ucrtbase.dll (the Universal C Runtime) So MediaPipe crashes with function 'free' not found The Fix:

ACheck if you're on Windows (os.name == 'nt') If yes, load the Windows C runtime (ucrtbase.dll) separately Grab the free function from there and attach it to the MediaPipe library object Now when MediaPipe looks for free(), it finds the one we attached In short: MediaPipe looks for free() in the wrong place on Windows. The fix just points it to the right place.

This is a bug in MediaPipe. They didn't account for how Windows handles C runtime functions differently from Linux/macOS.

2nd Problem Created 4 3 files and modified 4th one (Calibration.py, gaze_estimation.py, visualization.py and mediapipe_new_model.py but the calibration accuracy is very poor. Grouping these in a folder named "P1" and trying out other scripts for better accuracy)

Solved by creating a seperate "gaze_calibration.py", added calibration technique using two dots, one for optical axis (predicted gaze using center of the eye) the other one is for visual axis (actual gaze, user-specific), added accuracy measures but it's not working yet. Integrated pre-trained model: Blendshapes (face_blendshapes.tflite) [very useful for deepfake detection]. The traction needs alot of improvements but is working fine for now, going to add heart rate monitor

[Run gaze_calibration.py for now, I'm working on the folder structure aswell]