Breaking the accuracy-resource dilemma: a lightweight adaptive video inference enhancement
arXiv:2601.14568v2 Announce Type: replace-cross
Abstract: Existing video inference (VI) enhancement methods typically aim to improve performance by scaling up model sizes and employing sophisticated network architectures. While these approaches demons…