Compositional Benchmark Synthesis for Hierarchical Human Action Recognition

作者：Farnaz Soleimani (LISSI), Abdelghani Chibani (LISSI), Yacine Amirat (LISSI), Ghazaleh Khodabandelou (LISSI) · 单位：A benchmark-generation and evaluation framework that builds a four-level intention hierarchy from a flat, single-label action corpus., isolated clips, 1 label each, labeled episodes · 会议/期刊：arXiv preprint · 方向：cs.AI · 发布日期：2026-08-12