Say nothing for now, maybe later... My best result (mean error rate % normalized by inter_pupil) on 300W is as follow: Challenge Subset Common Subset Fullset 9.11 4.97 5.78