SHINE: Saliency-aware HIerarchical NEgative Ranking for Compositional Temporal Grounding

TL;DR

This paper introduces SHINE, a novel approach for compositional temporal grounding that leverages large language models for generating hard negative queries and employs a hierarchical saliency ranking strategy to improve model generalization.

Contribution

The paper proposes a new method combining GPT-3.5-Turbo generated negative queries with a coarse-to-fine saliency ranking to enhance compositional generalization in video moment retrieval.

Findings

Significant improvement on two challenging benchmarks.

Enhanced model robustness to subtle negative query differences.

Better saliency response accuracy in negative query scenarios.

Abstract

Temporal grounding, also known as video moment retrieval, aims at locating video segments corresponding to a given query sentence. The compositional nature of natural language enables the localization beyond predefined events, posing a certain challenge to the compositional generalizability of existing methods. Recent studies establish the correspondence between videos and queries through a decompose-reconstruct manner to achieve compositional generalization. However, they only consider dominant primitives and build negative queries through random sampling and recombination, resulting in semantically implausible negatives that hinder the models from learning rational compositions. In addition, recent DETR-based methods still underperform in compositional temporal grounding, showing irrational saliency responses when given negative queries that have subtle differences from positive queries. To address these limitations, we first propose a large language model-driven method for negative query construction, utilizing GPT-3.5-Turbo to generate semantically plausible hard negative queries. Subsequently, we introduce a coarse-to-fine saliency ranking strategy, which encourages the model to learn the multi-granularity semantic relationships between videos and hierarchical negative queries to boost compositional generalization. Extensive experiments on two challenging benchmarks validate the effectiveness and generalizability of our proposed method. Our code is available at https://github.com/zxccade/SHINE.