Two-Stream Consensus Network: Submission to HACS Challenge 2021 Weakly-Supervised Learning Track

TL;DR

This paper introduces a two-stream consensus network for weakly-supervised temporal action localization, achieving competitive results in the HACS Challenge 2021 by effectively combining RGB and optical flow data without fine-tuning feature extractors.

Contribution

The paper proposes a novel two-stream consensus network framework that generates pseudo ground truths for weakly-supervised learning in untrimmed videos.

Findings

Achieved 22.20% mAP on validation set

Ranked 2nd in HACS Challenge 2021

Effective fusion of RGB and optical flow modalities

Abstract

This technical report presents our solution to the HACS Temporal Action Localization Challenge 2021, Weakly-Supervised Learning Track. The goal of weakly-supervised temporal action localization is to temporally locate and classify action of interest in untrimmed videos given only video-level labels. We adopt the two-stream consensus network (TSCN) as the main framework in this challenge. The TSCN consists of a two-stream base model training procedure and a pseudo ground truth learning procedure. The base model training encourages the model to predict reliable predictions based on single modality (i.e., RGB or optical flow), based on the fusion of which a pseudo ground truth is generated and in turn used as supervision to train the base models. On the HACS v1.1.1 dataset, without fine-tuning the feature-extraction I3D models, our method achieves 22.20% on the validation set and 21.68% on the testing set in terms of average mAP. Our solution ranked the 2nd in this challenge, and we hope our method can serve as a baseline for future academic research.