SpotFast Networks with Memory Augmented Lateral Transformers for Lipreading

TL;DR

This paper introduces SpotFast networks with memory-augmented lateral transformers for word-level lipreading, achieving improved accuracy by combining a novel architecture with sequential learning enhancements.

Contribution

The paper proposes a new SpotFast network architecture with memory-augmented lateral transformers for lipreading, outperforming existing models on the LRW dataset.

Findings

Outperforms state-of-the-art lipreading models

Memory-augmented lateral transformers improve accuracy by 3.7%

Effective use of temporal window and all frames in lipreading

Abstract

This paper presents a novel deep learning architecture for word-level lipreading. Previous works suggest a potential for incorporating a pretrained deep 3D Convolutional Neural Networks as a front-end feature extractor. We introduce a SpotFast networks, a variant of the state-of-the-art SlowFast networks for action recognition, which utilizes a temporal window as a spot pathway and all frames as a fast pathway. We further incorporate memory augmented lateral transformers to learn sequential features for classification. We evaluate the proposed model on the LRW dataset. The experiments show that our proposed model outperforms various state-of-the-art models and incorporating the memory augmented lateral transformers makes a 3.7% improvement to the SpotFast networks.