Mamba Fusion: Learning Actions Through Questioning

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Dong, Zhikang, Beedu, Apoorva, Sheinkopf, Jason, Essa, Irfan
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913672212774912
author Dong, Zhikang
Beedu, Apoorva
Sheinkopf, Jason
Essa, Irfan
author_facet Dong, Zhikang
Beedu, Apoorva
Sheinkopf, Jason
Essa, Irfan
contents Video Language Models (VLMs) are crucial for generalizing across diverse tasks and using language cues to enhance learning. While transformer-based architectures have been the de facto in vision-language training, they face challenges like quadratic computational complexity, high GPU memory usage, and difficulty with long-term dependencies. To address these limitations, we introduce MambaVL, a novel model that leverages recent advancements in selective state space modality fusion to efficiently capture long-range dependencies and learn joint representations for vision and language data. MambaVL utilizes a shared state transition matrix across both modalities, allowing the model to capture information about actions from multiple perspectives within the scene. Furthermore, we propose a question-answering task that helps guide the model toward relevant cues. These questions provide critical information about actions, objects, and environmental context, leading to enhanced performance. As a result, MambaVL achieves state-of-the-art performance in action recognition on the Epic-Kitchens-100 dataset and outperforms baseline methods in action anticipation.
format Preprint
id arxiv_https___arxiv_org_abs_2409_11513
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Mamba Fusion: Learning Actions Through Questioning
Dong, Zhikang
Beedu, Apoorva
Sheinkopf, Jason
Essa, Irfan
Computer Vision and Pattern Recognition
Artificial Intelligence
Video Language Models (VLMs) are crucial for generalizing across diverse tasks and using language cues to enhance learning. While transformer-based architectures have been the de facto in vision-language training, they face challenges like quadratic computational complexity, high GPU memory usage, and difficulty with long-term dependencies. To address these limitations, we introduce MambaVL, a novel model that leverages recent advancements in selective state space modality fusion to efficiently capture long-range dependencies and learn joint representations for vision and language data. MambaVL utilizes a shared state transition matrix across both modalities, allowing the model to capture information about actions from multiple perspectives within the scene. Furthermore, we propose a question-answering task that helps guide the model toward relevant cues. These questions provide critical information about actions, objects, and environmental context, leading to enhanced performance. As a result, MambaVL achieves state-of-the-art performance in action recognition on the Epic-Kitchens-100 dataset and outperforms baseline methods in action anticipation.
title Mamba Fusion: Learning Actions Through Questioning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2409.11513