Leveraging a self-adaptive mean teacher model for semi-supervised multi-exposure image fusion.