arxivcs.CV2026-07-01
QCA: Query- and Content-Aware Keyframe Selection for Long Video Understanding
Jun Peng, Baiyang Song, Jie Li, Hui Li, Yiyi Zhou, Rongrong Ji, et al.
Video understanding is often plagued by severe temporal redundancy, where processing dense frame sequences is both semantically inefficient and computationally expensive. This challenge is further amplified when only a small subset of frames is truly relevant to the given query.…