Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Achieving 3D Attention via Triplet Squeeze and Excitation Block

Maan Alhazmi, Abdulrahman Altahhan

TL;DR

This paper introduces TripSE, a 3D attention block that merges Triplet Attention (TA) with Squeeze-and-Excitation (SE) to capture cross-dimensional dependencies and global channel importance. Four TripSE variants are developed and integrated into ResNet18, DenseNet, and ConvNeXt, achieving effective gains with minimal overhead, especially for ConvNeXt. Comprehensive experiments on CIFAR100, ImageNet, FER2013, and AffectNet demonstrate consistent improvements, including a new state-of-the-art 78.27% accuracy on FER2013 with ConvNeXt + TripSE. The work highlights the continued relevance of CNN-based architectures and attention mechanisms for vision tasks, particularly facial expression recognition, and provides a practical path to enhanced performance via 3D attention.

Abstract

The emergence of ConvNeXt and its variants has reaffirmed the conceptual and structural suitability of CNN-based models for vision tasks, re-establishing them as key players in image classification in general, and in facial expression recognition (FER) in particular. In this paper, we propose a new set of models that build on these advancements by incorporating a new set of attention mechanisms that combines Triplet attention with Squeeze-and-Excitation (TripSE) in four different variants. We demonstrate the effectiveness of these variants by applying them to the ResNet18, DenseNet and ConvNext architectures to validate their versatility and impact. Our study shows that incorporating a TripSE block in these CNN models boosts their performances, particularly for the ConvNeXt architecture, indicating its utility. We evaluate the proposed mechanisms and associated models across four datasets, namely CIFAR100, ImageNet, FER2013 and AffectNet datasets, where ConvNext with TripSE achieves state-of-the-art results with an accuracy of \textbf{78.27\%} on the popular FER2013 dataset, a new feat for this dataset.

Achieving 3D Attention via Triplet Squeeze and Excitation Block

TL;DR

This paper introduces TripSE, a 3D attention block that merges Triplet Attention (TA) with Squeeze-and-Excitation (SE) to capture cross-dimensional dependencies and global channel importance. Four TripSE variants are developed and integrated into ResNet18, DenseNet, and ConvNeXt, achieving effective gains with minimal overhead, especially for ConvNeXt. Comprehensive experiments on CIFAR100, ImageNet, FER2013, and AffectNet demonstrate consistent improvements, including a new state-of-the-art 78.27% accuracy on FER2013 with ConvNeXt + TripSE. The work highlights the continued relevance of CNN-based architectures and attention mechanisms for vision tasks, particularly facial expression recognition, and provides a practical path to enhanced performance via 3D attention.

Abstract

The emergence of ConvNeXt and its variants has reaffirmed the conceptual and structural suitability of CNN-based models for vision tasks, re-establishing them as key players in image classification in general, and in facial expression recognition (FER) in particular. In this paper, we propose a new set of models that build on these advancements by incorporating a new set of attention mechanisms that combines Triplet attention with Squeeze-and-Excitation (TripSE) in four different variants. We demonstrate the effectiveness of these variants by applying them to the ResNet18, DenseNet and ConvNext architectures to validate their versatility and impact. Our study shows that incorporating a TripSE block in these CNN models boosts their performances, particularly for the ConvNeXt architecture, indicating its utility. We evaluate the proposed mechanisms and associated models across four datasets, namely CIFAR100, ImageNet, FER2013 and AffectNet datasets, where ConvNext with TripSE achieves state-of-the-art results with an accuracy of \textbf{78.27\%} on the popular FER2013 dataset, a new feat for this dataset.
Paper Structure (18 sections, 4 figures, 3 tables)

This paper contains 18 sections, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Attention Mechanisms in Computer Vision
  4. Models and Attention Mechanism for FER
  5. Methodology
  6. Triplet Attention (TA)
  7. Squeeze-and-Excitation (SE)
  8. Triplet Squeeze-and-Excitation Attention (TripSE)
  9. Variants of the TripSE Block
  10. Incorporating the TripSE block
  11. Experiments
  12. Datasets
  13. Implementation Details
  14. Results
  15. Initial Study
  16. ...and 3 more sections

Figures (4)

  • Figure 1: ConvNext, DenseNet, and ResNet architectures incorporated with the TripSE block.
  • Figure 2: Triplet Attention with Squeeze-and-Excitation Blocks 1 (TripSE1) and 2 (TripSE2). TripSE1 is achieved by activating the sky-blue block and deactivating the yellow blocks, where TripSE2 is achieved visa-versa.
  • Figure 3: Triplet Attention with Squeeze-and-Excitation Block 3 (TripSE 3).
  • Figure 4: Triplet Attention with Squeeze-and-Excitation Block 4 (TripSE 4).