HERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training

TL;DR

HERO is a hierarchical video+language pre-training framework that captures local and global context through novel tasks, achieving state-of-the-art results on multiple video understanding benchmarks.

Contribution

The paper introduces HERO, a hierarchical encoder with new pre-training tasks for improved video+language understanding and benchmarks.

Findings

Achieves new state-of-the-art on multiple video understanding benchmarks.

Introduces two new challenging benchmarks How2QA and How2R.

Demonstrates effectiveness of hierarchical encoding and novel pre-training tasks.

Abstract

We present HERO, a novel framework for large-scale video+language omni-representation learning. HERO encodes multimodal inputs in a hierarchical structure, where local context of a video frame is captured by a Cross-modal Transformer via multimodal fusion, and global video context is captured by a Temporal Transformer. In addition to standard Masked Language Modeling (MLM) and Masked Frame Modeling (MFM) objectives, we design two new pre-training tasks: (i) Video-Subtitle Matching (VSM), where the model predicts both global and local temporal alignment; and (ii) Frame Order Modeling (FOM), where the model predicts the right order of shuffled video frames. HERO is jointly trained on HowTo100M and large-scale TV datasets to gain deep understanding of complex social dynamics with multi-character interactions. Comprehensive experiments demonstrate that HERO achieves new state of the art on multiple benchmarks over Text-based Video/Video-moment Retrieval, Video Question Answering (QA), Video-and-language Inference and Video Captioning tasks across different domains. We also introduce two new challenging benchmarks How2QA and How2R for Video QA and Retrieval, collected from diverse video content over multimodalities.