Domain generalization using action sequences for egocentric action recognition

Download PDF
Open Webpage

Amirshayan Nasirimajd, Chiara Plizzari, Simone Alberto Peirone, Marco Ciccone, Giuseppe Averta, Barbara Caputo

Published: 2025, Last Modified: 10 Nov 2025Pattern Recognit. Lett. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We investigate the role of action sequences in Egocentric Action Recognition (EAR).•We present SeqDG, a novel model to exploit sequences to improve generalization in EAR.•SeqDG deploys a masked reconstruction objective to use context from action sequences.•SeqDG achieves SOTA performance on the EPIC-Kitchens-100 and EGTEA benchmarks.