Abstract Scaling video predictive models to higher resolutions typically requires retraining from scratch, incurring substantial computationalcosts. We present Walsh-Hadamard Spec...