Saved in:
Bibliographic Details
Main Authors: Wang, Haoqi, Zhang, Tong, Salzmann, Mathieu
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.16826
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911965178232832
author Wang, Haoqi
Zhang, Tong
Salzmann, Mathieu
author_facet Wang, Haoqi
Zhang, Tong
Salzmann, Mathieu
contents Vision Transformer models trained on large-scale datasets, although effective, often exhibit artifacts in the patch token they extract. While such defects can be alleviated by re-training the entire model with additional classification tokens, the underlying reasons for the presence of these tokens remain unclear. In this paper, we conduct a thorough investigation of this phenomenon, combining theoretical analysis with empirical observations. Our findings reveal that these artifacts originate from the pre-trained network itself, specifically stemming from the leading left singular vector of the network's weights. Furthermore, to mitigate these defects, we propose a novel fine-tuning smooth regularization that rectifies structural deficiencies using only a small dataset, thereby avoiding the need for complete re-training. We validate our method on various downstream tasks, including unsupervised segmentation, classification, supervised segmentation, and depth estimation, demonstrating its effectiveness in improving model performance. Codes and checkpoints are available at https://github.com/haoqiwang/sinder.
format Preprint
id arxiv_https___arxiv_org_abs_2407_16826
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SINDER: Repairing the Singular Defects of DINOv2
Wang, Haoqi
Zhang, Tong
Salzmann, Mathieu
Computer Vision and Pattern Recognition
Vision Transformer models trained on large-scale datasets, although effective, often exhibit artifacts in the patch token they extract. While such defects can be alleviated by re-training the entire model with additional classification tokens, the underlying reasons for the presence of these tokens remain unclear. In this paper, we conduct a thorough investigation of this phenomenon, combining theoretical analysis with empirical observations. Our findings reveal that these artifacts originate from the pre-trained network itself, specifically stemming from the leading left singular vector of the network's weights. Furthermore, to mitigate these defects, we propose a novel fine-tuning smooth regularization that rectifies structural deficiencies using only a small dataset, thereby avoiding the need for complete re-training. We validate our method on various downstream tasks, including unsupervised segmentation, classification, supervised segmentation, and depth estimation, demonstrating its effectiveness in improving model performance. Codes and checkpoints are available at https://github.com/haoqiwang/sinder.
title SINDER: Repairing the Singular Defects of DINOv2
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2407.16826