FBK-HUPBA Submission to the EPIC-Kitchens 2019 Action Recognition Challenge

TL;DR

This paper details the FBK-HUPBA team's approach to the EPIC-Kitchens 2019 challenge, using CNN-LSTA and HF-TSN models in an ensemble to achieve competitive action recognition accuracy.

Contribution

The paper introduces an ensemble of CNN-LSTA and HF-TSN models for action recognition in egocentric videos, demonstrating improved accuracy on the EPIC-Kitchens benchmark.

Findings

Achieved 35.54% top-1 accuracy on S1 setting.

Achieved 20.25% top-1 accuracy on S2 setting.

Ensemble of CNN-LSTA and HF-TSN models outperforms individual models.

Abstract

In this report we describe the technical details of our submission to the EPIC-Kitchens 2019 action recognition challenge. To participate in the challenge we have developed a number of CNN-LSTA [3] and HF-TSN [2] variants, and submitted predictions from an ensemble compiled out of these two model families. Our submission, visible on the public leaderboard with team name FBK-HUPBA, achieved a top-1 action recognition accuracy of 35.54% on S1 setting, and 20.25% on S2 setting.