RUS  ENG
Full version
JOURNALS // Computer Optics // Archive

Computer Optics, 2023 Volume 47, Issue 4, Pages 637–649 (Mi co1165)

This article is cited in 2 papers

IMAGE PROCESSING, PATTERN RECOGNITION

Mutual modality learning for video action classification

S. A. Komkovab, M. D. Dzabraevab, A. A. Petiushkoab

a Lomonosov Moscow State University
b Huawei Moscow Research Center, 121099, Russia, Moscow, Smolenskaya ploshchad 7–9

Abstract: The construction of models for video action classification progresses rapidly. However, the performance of those models can still be easily improved by ensembling with the same models trained on different modalities (e.g. Optical flow). Unfortunately, it is computationally expensive to use several modalities during inference. Recent works examine the ways to integrate advantages of multi-modality into a single RGB-model. Yet, there is still room for improvement. In this paper, we explore various methods to embed the ensemble power into a single model. We show that proper initialization, as well as mutual modality learning, enhances single-modality models. As a result, we achieve state-of-the-art results in the Something-Something-v2 benchmark.

Keywords: video recognition, video action classification, video labeling, mutual learning, optical flow

Received: 13.01.2023
Accepted: 29.03.2023

Language: English

DOI: 10.18287/2412-6179-CO-1277



© Steklov Math. Inst. of RAS, 2025