Beyond Grids: Exploring Elastic Input Sampling for Vision Transformers

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Pardyl, Adam, Kurzejamski, Grzegorz, Olszewski, Jan, Trzciński, Tomasz, Zieliński, Bartosz
Formato: Preprint
Publicado: 2023
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918164286144512
author Pardyl, Adam
Kurzejamski, Grzegorz
Olszewski, Jan
Trzciński, Tomasz
Zieliński, Bartosz
author_facet Pardyl, Adam
Kurzejamski, Grzegorz
Olszewski, Jan
Trzciński, Tomasz
Zieliński, Bartosz
contents Vision transformers have excelled in various computer vision tasks but mostly rely on rigid input sampling using a fixed-size grid of patches. It limits their applicability in real-world problems, such as active visual exploration, where patches have various scales and positions. Our paper addresses this limitation by formalizing the concept of input elasticity for vision transformers and introducing an evaluation protocol for measuring this elasticity. Moreover, we propose modifications to the transformer architecture and training regime, which increase its elasticity. Through extensive experimentation, we spotlight opportunities and challenges associated with such architecture.
format Preprint
id arxiv_https___arxiv_org_abs_2309_13353
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Beyond Grids: Exploring Elastic Input Sampling for Vision Transformers
Pardyl, Adam
Kurzejamski, Grzegorz
Olszewski, Jan
Trzciński, Tomasz
Zieliński, Bartosz
Computer Vision and Pattern Recognition
Vision transformers have excelled in various computer vision tasks but mostly rely on rigid input sampling using a fixed-size grid of patches. It limits their applicability in real-world problems, such as active visual exploration, where patches have various scales and positions. Our paper addresses this limitation by formalizing the concept of input elasticity for vision transformers and introducing an evaluation protocol for measuring this elasticity. Moreover, we propose modifications to the transformer architecture and training regime, which increase its elasticity. Through extensive experimentation, we spotlight opportunities and challenges associated with such architecture.
title Beyond Grids: Exploring Elastic Input Sampling for Vision Transformers
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2309.13353