FBK-HUPBA Submission to the EPIC-Kitchens Action Recognition 2020 Challenge

TL;DR

This paper details the FBK-HUPBA team's submission to the EPIC-Kitchens 2020 challenge, utilizing advanced spatio-temporal models and ensemble techniques to achieve competitive action recognition accuracy using only RGB data.

Contribution

The paper introduces an ensemble of GSM and EgoACO models with various backbones and pre-training for improved action recognition in egocentric videos.

Findings

Achieved 40.0% top-1 accuracy on S1 setting

Achieved 25.71% top-1 accuracy on S2 setting

Demonstrated effectiveness of ensemble of GSM and EgoACO models

Abstract

In this report we describe the technical details of our submission to the EPIC-Kitchens Action Recognition 2020 Challenge. To participate in the challenge we deployed spatio-temporal feature extraction and aggregation models we have developed recently: Gate-Shift Module (GSM) [1] and EgoACO, an extension of Long Short-Term Attention (LSTA) [2]. We design an ensemble of GSM and EgoACO model families with different backbones and pre-training to generate the prediction scores. Our submission, visible on the public leaderboard with team name FBK-HUPBA, achieved a top-1 action recognition accuracy of 40.0% on S1 setting, and 25.71% on S2 setting, using only RGB.