Extracting temporal features into a spatial domain using autoencoders for sperm video analysis