Progressive Boundary Refinement Network for Temporal Action Detection

• DOI: 10.1609/aaai.v34i07.6829
• Publication Date: 2020-06-29T18:34:40Z
• Authors: Qinying Liu; Zilei Wang
• Keywords: Vagueness; Feature (linguistics)

Temporal action detection is a challenging task due to vagueness of boundaries. To tackle this issue, we propose an end-to-end progressive boundary refinement network (PBRNet) in paper. PBRNet belongs the family one-stage detectors and equipped with three cascaded modules for localizing more precisely. Specifically, mainly consists coarse pyramidal detection, refined fine-grained detection. The first two build feature pyramids perform anchor-based third one explores frame-level features refine boundaries each instance. In fined-grained module, classification branches are proposed augment update confidence scores instances. Evidently, integrates methods. We experimentally evaluate comprehensively investigate effect main components. results show achieves state-of-the-art performances on popular benchmarks: THUMOS'14 ActivityNet, meanwhile possesses high inference speed.