Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Pătrăucean, Viorica, He, Xu Owen, Heyward, Joseph, Zhang, Chuhan, Sajjadi, Mehdi S. M., Muraru, George-Cristian, Zholus, Artem, Karami, Mahdi, Goroshin, Ross, Chen, Yutian, Osindero, Simon, Carreira, João, Pascanu, Razvan
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2412.14294
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914330133397504
author Pătrăucean, Viorica
He, Xu Owen
Heyward, Joseph
Zhang, Chuhan
Sajjadi, Mehdi S. M.
Muraru, George-Cristian
Zholus, Artem
Karami, Mahdi
Goroshin, Ross
Chen, Yutian
Osindero, Simon
Carreira, João
Pascanu, Razvan
author_facet Pătrăucean, Viorica
He, Xu Owen
Heyward, Joseph
Zhang, Chuhan
Sajjadi, Mehdi S. M.
Muraru, George-Cristian
Zholus, Artem
Karami, Mahdi
Goroshin, Ross
Chen, Yutian
Osindero, Simon
Carreira, João
Pascanu, Razvan
contents We propose a novel block for \emph{causal} video modelling. It relies on a time-space-channel factorisation with dedicated blocks for each dimension: gated linear recurrent units (LRUs) perform information mixing over time, self-attention layers perform mixing over space, and MLPs over channels. The resulting architecture \emph{TRecViT} is causal and shows strong performance on sparse and dense tasks, trained in supervised or self-supervised regimes, being the first causal video model in the state-space models family. Notably, our model outperforms or is on par with the popular (non-causal) ViViT-L model on large scale video datasets (SSv2, Kinetics400), while having $3\times$ less parameters, $12\times$ smaller memory footprint, and $5\times$ lower FLOPs count than the full self-attention ViViT, with an inference throughput of about 300 frames per second, running comfortably in real-time. When compared with causal transformer-based models (TSM, RViT) and other recurrent models like LSTM, TRecViT obtains state-of-the-art results on the challenging SSv2 dataset. Code and checkpoints are available online https://github.com/google-deepmind/trecvit.
format Preprint
id arxiv_https___arxiv_org_abs_2412_14294
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle TRecViT: A Recurrent Video Transformer
Pătrăucean, Viorica
He, Xu Owen
Heyward, Joseph
Zhang, Chuhan
Sajjadi, Mehdi S. M.
Muraru, George-Cristian
Zholus, Artem
Karami, Mahdi
Goroshin, Ross
Chen, Yutian
Osindero, Simon
Carreira, João
Pascanu, Razvan
Computer Vision and Pattern Recognition
Machine Learning
We propose a novel block for \emph{causal} video modelling. It relies on a time-space-channel factorisation with dedicated blocks for each dimension: gated linear recurrent units (LRUs) perform information mixing over time, self-attention layers perform mixing over space, and MLPs over channels. The resulting architecture \emph{TRecViT} is causal and shows strong performance on sparse and dense tasks, trained in supervised or self-supervised regimes, being the first causal video model in the state-space models family. Notably, our model outperforms or is on par with the popular (non-causal) ViViT-L model on large scale video datasets (SSv2, Kinetics400), while having $3\times$ less parameters, $12\times$ smaller memory footprint, and $5\times$ lower FLOPs count than the full self-attention ViViT, with an inference throughput of about 300 frames per second, running comfortably in real-time. When compared with causal transformer-based models (TSM, RViT) and other recurrent models like LSTM, TRecViT obtains state-of-the-art results on the challenging SSv2 dataset. Code and checkpoints are available online https://github.com/google-deepmind/trecvit.
title TRecViT: A Recurrent Video Transformer
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2412.14294