Table 1: Evaluation of semi-supervised learning with respect to number of sub-dataset. Each value denotes the angular error between the predicted and ground truth gaze vector in degrees.

Method	3k	5k	10k	20k	50k
Baseline	10.69	8.89	7.42	6.47	5.60
RAT [15]	9.47	8.25	6.83	6.34	5.89
Ours	9.40	8.20	6.68	6.09	5.48