self-supervision

We show that input-aware parameter adaptation via meta-learning can greatly improve state-of-the-art video frame interpolation models.