Scaling error in landmarks.py

[HarryWindsor]

I am trying to run them demo. Camera Calibration and fine-tuning worked, but the gaze prediction on the screen do not work because the detected face landmarks are scaled in a wired way.

I only ran landmarks.py and can confirm that the error is also present there, see example blow. Did anybody else experience these issues?

Thank you for your help.

[parallelepipede]

I had the same problem. I modified the landmarks.py file to no longer use the HRNET model for face detection. I plugged instead the mediapipe face mesh and with some corrections (transform the 468 points of mediapipe for the 68 of dlib among others), it works and it is more fluid on my computer without gpu.