Temporal Context Modeling Network with Local-Global Complementary Architecture for Temporal Proposal Generation

Temporal Action Proposal Generation (TAPG) is a promising but challenging task with wide range of practical applications. Although state-of-the-art methods have made significant progress in TAPG, most ignore the impact temporal scales action and lack exploitation effective boundary contexts. In this paper, we propose simple unified framework named Context Modeling Network (TCMNet) that generates proposals. TCMNet innovatively uses convolutional filters different dilation rates to address scale issue. Specifically, contains BaseNet dilated convolutions (DBNet), an Completeness Module (ACM), Boundary Generator (TBG). The DBNet aims model information. It handles input video features through layers outputs feature sequence as ACM TBG. evaluate confidence scores densely distributed TBG designed enrich context instance. can generate boundaries high precision recall local–global complementary structure. We conduct comprehensive evaluations on two benchmarks: ActivityNet-1.3 THUMOS14. Extensive experiments demonstrate effectiveness proposed tasks proposal generation detection.