goatman/metahuman-gaze-prediction

#Extract and normalize the coordinates (dodgy version for testing) def get_coords_metahuman(file: Path): im_id, character, xcoord, ycoord, xsize, ysize = file.name.split('.jpg')[:-1][0].split('_')

xcoord, ycoord, xsize, ysize = float(xcoord), float(ycoord), float(xsize), float(ysize)
base_screensize = tensor([46.49, 26.15]) # generic width and height measurement in cms given by gpt4 as a likely mean screen size

normalized_screensize = tensor([xsize, ysize])/base_screensize


x = (xcoord)/xsize
y = (ycoord)/ysize

# normalize to range -0.5, 0.5


return tensor([x, y])