Robust Bird's Eye View Segmentation by Adapting DINOv2

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Barın, Merve Rabia, Aydemir, Görkay, Güney, Fatma
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916396406931456
author Barın, Merve Rabia
Aydemir, Görkay
Güney, Fatma
author_facet Barın, Merve Rabia
Aydemir, Görkay
Güney, Fatma
contents Extracting a Bird's Eye View (BEV) representation from multiple camera images offers a cost-effective, scalable alternative to LIDAR-based solutions in autonomous driving. However, the performance of the existing BEV methods drops significantly under various corruptions such as brightness and weather changes or camera failures. To improve the robustness of BEV perception, we propose to adapt a large vision foundational model, DINOv2, to BEV estimation using Low Rank Adaptation (LoRA). Our approach builds on the strong representation space of DINOv2 by adapting it to the BEV task in a state-of-the-art framework, SimpleBEV. Our experiments show increased robustness of BEV perception under various corruptions, with increasing gains from scaling up the model and the input resolution. We also showcase the effectiveness of the adapted representations in terms of fewer learnable parameters and faster convergence during training.
format Preprint
id arxiv_https___arxiv_org_abs_2409_10228
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Robust Bird's Eye View Segmentation by Adapting DINOv2
Barın, Merve Rabia
Aydemir, Görkay
Güney, Fatma
Computer Vision and Pattern Recognition
Extracting a Bird's Eye View (BEV) representation from multiple camera images offers a cost-effective, scalable alternative to LIDAR-based solutions in autonomous driving. However, the performance of the existing BEV methods drops significantly under various corruptions such as brightness and weather changes or camera failures. To improve the robustness of BEV perception, we propose to adapt a large vision foundational model, DINOv2, to BEV estimation using Low Rank Adaptation (LoRA). Our approach builds on the strong representation space of DINOv2 by adapting it to the BEV task in a state-of-the-art framework, SimpleBEV. Our experiments show increased robustness of BEV perception under various corruptions, with increasing gains from scaling up the model and the input resolution. We also showcase the effectiveness of the adapted representations in terms of fewer learnable parameters and faster convergence during training.
title Robust Bird's Eye View Segmentation by Adapting DINOv2
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2409.10228