nano-JEPA: Democratizing Video Understanding with Personal Computers

The Video Joint Embedding Predictive Architecture (V-JEPA) has shown great promise in self-supervised video representation learning. However, its substantial computational demands, often necessitates powerful GPU clusters, limit accessibility for many researchers. We introduce nano-JEPA, a streamlin...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Rostagno, Adrián, Iparraguirre, Javier, Ermantraut, Joel, Tobio, Lucas, Foissac, Segundo, Aggio, Santiago, Friedrich, Guillermo Rodolfo
Formato:	Objeto de conferencia
Lenguaje:	Inglés
Publicado:	2024
Materias:	Ciencias Informáticas feature prediction unsupervised learning visual representations video joint-embedding predictive architecture
Acceso en línea:	http://sedici.unlp.edu.ar/handle/10915/176281
Aporte de:	SEDICI (UNLP) de Universidad Nacional de La Plata

id	I19-R120-10915-176281
record_format	dspace
spelling	I19-R120-10915-1762812025-02-07T20:05:00Z http://sedici.unlp.edu.ar/handle/10915/176281 nano-JEPA: Democratizing Video Understanding with Personal Computers Rostagno, Adrián Iparraguirre, Javier Ermantraut, Joel Tobio, Lucas Foissac, Segundo Aggio, Santiago Friedrich, Guillermo Rodolfo 2024-10 2024 2025-02-07T16:57:48Z en Ciencias Informáticas feature prediction unsupervised learning visual representations video joint-embedding predictive architecture The Video Joint Embedding Predictive Architecture (V-JEPA) has shown great promise in self-supervised video representation learning. However, its substantial computational demands, often necessitates powerful GPU clusters, limit accessibility for many researchers. We introduce nano-JEPA, a streamlined adaptation of V-JEPA designed to run efficiently on resource-constrained personal computers, even those with only CPUs. Additionally, we present the nano-datasets repository, facilitating the creation of manageable subsets from large public video datasets. Our work aims to democratize research in this field, enabling broader participation and experimentation with V-JEPA-like models. We demonstrate that nano-JEPA, trained on smaller datasets and hardware, can still achieve reasonable performance on downstream tasks, opening doors for further exploration and innovation. Red de Universidades con Carreras en Informática Objeto de conferencia Objeto de conferencia http://creativecommons.org/licenses/by-nc-sa/4.0/ Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) application/pdf 94-103
institution	Universidad Nacional de La Plata
institution_str	I-19
repository_str	R-120
collection	SEDICI (UNLP)
language	Inglés
topic	Ciencias Informáticas feature prediction unsupervised learning visual representations video joint-embedding predictive architecture
spellingShingle	Ciencias Informáticas feature prediction unsupervised learning visual representations video joint-embedding predictive architecture Rostagno, Adrián Iparraguirre, Javier Ermantraut, Joel Tobio, Lucas Foissac, Segundo Aggio, Santiago Friedrich, Guillermo Rodolfo nano-JEPA: Democratizing Video Understanding with Personal Computers
topic_facet	Ciencias Informáticas feature prediction unsupervised learning visual representations video joint-embedding predictive architecture
description	The Video Joint Embedding Predictive Architecture (V-JEPA) has shown great promise in self-supervised video representation learning. However, its substantial computational demands, often necessitates powerful GPU clusters, limit accessibility for many researchers. We introduce nano-JEPA, a streamlined adaptation of V-JEPA designed to run efficiently on resource-constrained personal computers, even those with only CPUs. Additionally, we present the nano-datasets repository, facilitating the creation of manageable subsets from large public video datasets. Our work aims to democratize research in this field, enabling broader participation and experimentation with V-JEPA-like models. We demonstrate that nano-JEPA, trained on smaller datasets and hardware, can still achieve reasonable performance on downstream tasks, opening doors for further exploration and innovation.
format	Objeto de conferencia Objeto de conferencia
author	Rostagno, Adrián Iparraguirre, Javier Ermantraut, Joel Tobio, Lucas Foissac, Segundo Aggio, Santiago Friedrich, Guillermo Rodolfo
author_facet	Rostagno, Adrián Iparraguirre, Javier Ermantraut, Joel Tobio, Lucas Foissac, Segundo Aggio, Santiago Friedrich, Guillermo Rodolfo
author_sort	Rostagno, Adrián
title	nano-JEPA: Democratizing Video Understanding with Personal Computers
title_short	nano-JEPA: Democratizing Video Understanding with Personal Computers
title_full	nano-JEPA: Democratizing Video Understanding with Personal Computers
title_fullStr	nano-JEPA: Democratizing Video Understanding with Personal Computers
title_full_unstemmed	nano-JEPA: Democratizing Video Understanding with Personal Computers
title_sort	nano-jepa: democratizing video understanding with personal computers
publishDate	2024
url	http://sedici.unlp.edu.ar/handle/10915/176281
work_keys_str_mv	AT rostagnoadrian nanojepademocratizingvideounderstandingwithpersonalcomputers AT iparraguirrejavier nanojepademocratizingvideounderstandingwithpersonalcomputers AT ermantrautjoel nanojepademocratizingvideounderstandingwithpersonalcomputers AT tobiolucas nanojepademocratizingvideounderstandingwithpersonalcomputers AT foissacsegundo nanojepademocratizingvideounderstandingwithpersonalcomputers AT aggiosantiago nanojepademocratizingvideounderstandingwithpersonalcomputers AT friedrichguillermorodolfo nanojepademocratizingvideounderstandingwithpersonalcomputers
_version_	1845116777788342272

nano-JEPA: Democratizing Video Understanding with Personal Computers

Ejemplares similares