QSMT-net: A query-sensitive proposal and multi-temporal-span matching network for video grounding

Qingqing Wu, Lijun Guo, Rong Zhang, Jiangbo Qian, Shangce Gao

Abstract: Highlights•We designed a query-sensitive proposal generation strategy and dynamically generates candidate proposals through a constructed learnable pooling module.•We developed a multi-temporal-span matching network that simulated the matching between candidate proposals and queries across various temporal perspectives.•Our approach outperformed state-of-the-art methods on three challenging video localization benchmarks.