Event Transformer+. A multi-purpose solution for efficient event data processing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sabater, Alberto, Montesano, Luis, Murillo, Ana C.
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911762608029696
author Sabater, Alberto
Montesano, Luis
Murillo, Ana C.
author_facet Sabater, Alberto
Montesano, Luis
Murillo, Ana C.
contents Event cameras record sparse illumination changes with high temporal resolution and high dynamic range. Thanks to their sparse recording and low consumption, they are increasingly used in applications such as AR/VR and autonomous driving. Current topperforming methods often ignore specific event-data properties, leading to the development of generic but computationally expensive algorithms, while event-aware methods do not perform as well. We propose Event Transformer+, that improves our seminal work EvT with a refined patch-based event representation and a more robust backbone to achieve more accurate results, while still benefiting from event-data sparsity to increase its efficiency. Additionally, we show how our system can work with different data modalities and propose specific output heads, for event-stream classification (i.e. action recognition) and per-pixel predictions (dense depth estimation). Evaluation results show better performance to the state-of-the-art while requiring minimal computation resources, both on GPU and CPU.
format Preprint
id arxiv_https___arxiv_org_abs_2211_12222
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Event Transformer+. A multi-purpose solution for efficient event data processing
Sabater, Alberto
Montesano, Luis
Murillo, Ana C.
Computer Vision and Pattern Recognition
Event cameras record sparse illumination changes with high temporal resolution and high dynamic range. Thanks to their sparse recording and low consumption, they are increasingly used in applications such as AR/VR and autonomous driving. Current topperforming methods often ignore specific event-data properties, leading to the development of generic but computationally expensive algorithms, while event-aware methods do not perform as well. We propose Event Transformer+, that improves our seminal work EvT with a refined patch-based event representation and a more robust backbone to achieve more accurate results, while still benefiting from event-data sparsity to increase its efficiency. Additionally, we show how our system can work with different data modalities and propose specific output heads, for event-stream classification (i.e. action recognition) and per-pixel predictions (dense depth estimation). Evaluation results show better performance to the state-of-the-art while requiring minimal computation resources, both on GPU and CPU.
title Event Transformer+. A multi-purpose solution for efficient event data processing
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2211.12222