Skip to yearly menu bar Skip to main content


Poster

COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning

Simon Ging · Mohammadreza Zolfaghari · Hamed Pirsiavash · Thomas Brox

Poster Session 2 #630

Abstract:

Many real-world video-text tasks involve different levels of granularity, such as frames and words, clip and sentences or videos and paragraphs, each with distinct semantics. In this paper, we propose a Cooperative hierarchical Transformer (COOT) to leverage this hierarchy information and model the interactions between different levels of granularity and different modalities. The method consists of three major components: an attention-aware feature aggregation layer, which leverages the local temporal context (intra-level, e.g., within a clip), a contextual transformer to learn the interactions between low-level and high-level semantics (inter-level, e.g. clip-video, sentence-paragraph), and a cross-modal cycle-consistency loss to connect video and text. The resulting method compares favorably to the state of the art on several benchmarks while having few parameters.

Chat is not available.