Coarse to Fine: Video Retrieval before Moment Localization

Abstract

The current state-of-the-art methods for video corpus moment retrieval (VCMR) often use similarity-based feature alignment approach for the sake of convenience and speed. However, late fusion methods like cosine similarity alignment are unable to make full use of the information from both query texts and videos. In this paper, we combine feature alignment with feature fusion to promote the performance on VCMR.

References in corpus (4)

• HERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training
• ConSERT: A Contrastive Framework for Self-Supervised Sentence Representation Transfer
• VALUE: A Multi-Task Benchmark for Video-and-Language Understanding Evaluation
• A Hierarchical Multi-Modal Encoder for Moment Localization in Video Corpus