Saved in:
Bibliographic Details
Main Authors: Ouyang, Yuanbing, Liang, Yizhuo, Li, Qingpeng, Guo, Xinfei, Luo, Yiming, Wu, Di, Wang, Hao, Pan, Yushan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.17996
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912346388037632
author Ouyang, Yuanbing
Liang, Yizhuo
Li, Qingpeng
Guo, Xinfei
Luo, Yiming
Wu, Di
Wang, Hao
Pan, Yushan
author_facet Ouyang, Yuanbing
Liang, Yizhuo
Li, Qingpeng
Guo, Xinfei
Luo, Yiming
Wu, Di
Wang, Hao
Pan, Yushan
contents Vision Transformers (ViTs) excel in semantic segmentation but demand significant computation, posing challenges for deployment on resource-constrained devices. Existing token pruning methods often overlook fundamental visual data characteristics. This study introduces 'LVTP', a progressive token pruning framework guided by multi-scale Tsallis entropy and low-level visual features with twice clustering. It integrates high-level semantics and basic visual attributes for precise segmentation. A novel dynamic scoring mechanism using multi-scale Tsallis entropy weighting overcomes limitations of traditional single-parameter entropy. The framework also incorporates low-level feature analysis to preserve critical edge information while optimizing computational cost. As a plug-and-play module, it requires no architectural changes or additional training. Evaluations across multiple datasets show 20%-45% computational reductions with negligible performance loss, outperforming existing methods in balancing cost and accuracy, especially in complex edge regions.
format Preprint
id arxiv_https___arxiv_org_abs_2504_17996
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Back to Fundamentals: Low-Level Visual Features Guided Progressive Token Pruning
Ouyang, Yuanbing
Liang, Yizhuo
Li, Qingpeng
Guo, Xinfei
Luo, Yiming
Wu, Di
Wang, Hao
Pan, Yushan
Computer Vision and Pattern Recognition
Vision Transformers (ViTs) excel in semantic segmentation but demand significant computation, posing challenges for deployment on resource-constrained devices. Existing token pruning methods often overlook fundamental visual data characteristics. This study introduces 'LVTP', a progressive token pruning framework guided by multi-scale Tsallis entropy and low-level visual features with twice clustering. It integrates high-level semantics and basic visual attributes for precise segmentation. A novel dynamic scoring mechanism using multi-scale Tsallis entropy weighting overcomes limitations of traditional single-parameter entropy. The framework also incorporates low-level feature analysis to preserve critical edge information while optimizing computational cost. As a plug-and-play module, it requires no architectural changes or additional training. Evaluations across multiple datasets show 20%-45% computational reductions with negligible performance loss, outperforming existing methods in balancing cost and accuracy, especially in complex edge regions.
title Back to Fundamentals: Low-Level Visual Features Guided Progressive Token Pruning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.17996