Mutual Attention

Mutual attention mechanisms in computer vision aim to improve model performance by enabling different parts of an input (e.g., image patches, hand and object features, visual and textual information) to interact and learn from each other in a bidirectional manner. Current research focuses on applying mutual attention within various architectures, including Vision Transformers and graph neural networks, to enhance tasks such as few-shot learning, referring image segmentation, and 3D hand-object pose estimation. These advancements lead to more accurate and efficient models, particularly in scenarios with limited data or complex interactions between different modalities, impacting fields like robotics and human-computer interaction.

Papers

May 6, 2024

Intra-task Mutual Attention based Vision Transformer for Few-Shot Learning
Weihao Jiang, Chang Liu, Kun He
LeArning Abstract Vision Transformer Shot Learning Pre Trained Vision Transformer Intra Class Shot Classification Benchmark Mutual Attention

November 27, 2023

RISAM: Referring Image Segmentation via Mutual-Aware Attention Features
Mengxi Zhang, Yiming Liu, Xiangjun Yin, Huanjing Yue, Jingyu Yang
Image Segmentation Visual Context Consistent Segmentation Feature Wise Attention Mutual Attention

November 16, 2022

Interacting Hand-Object Pose Estimation via Dense Mutual Attention
Rong Wang, Wei Mao, Hongdong Li
Hand Pose Estimation Human Hand Plausible Pose Mutual Attention

Mutual Attention

Papers

Intra-task Mutual Attention based Vision Transformer for Few-Shot Learning

RISAM: Referring Image Segmentation via Mutual-Aware Attention Features

Interacting Hand-Object Pose Estimation via Dense Mutual Attention