1 paper
Ziyue Zeng, Xun Su, Haoyuan Liu +3
At ultra-low bitrates, high-fidelity reconstruction requires sampling plausible videos from the posterior rather than regressing to oversmoothed conditional means. We propose Gener…