ScopeViT: Scale-Aware Vision Transformer

计算机科学 人工智能 计算机视觉 变压器 工程类 电气工程 电压
作者
Xuesong Nie,Haoyuan Jin,Yunfeng Yan,Xi Chen,Zhihang Zhu,Donglian Qi
出处
期刊:Pattern Recognition [Elsevier BV]
卷期号:153: 110470-110470 被引量:8
标识
DOI:10.1016/j.patcog.2024.110470
摘要

Multi-scale features are essential for various vision tasks, such as classification, detection, and segmentation. Although Vision Transformers (ViTs) show remarkable success in capturing global features within an image, how to leverage multi-scale features in Transformers is not well explored. This paper proposes a scale-aware vision Transformer called ScopeViT that efficiently captures multi-granularity representations. Two novel attention with lightweight computation are introduced: Multi-Scale Self-Attention (MSSA) and Global-Scale Dilated Attention (GSDA). MSSA embeds visual tokens with different receptive fields into distinct attention heads, allowing the model to perceive various scales across the network. GSDA enhances model understanding of the global context through token-dilation operation, which reduces the number of tokens involved in attention computations. This dual attention method enables ScopeViT to "see" various scales throughout the entire network and effectively learn inter-object relationships, reducing heavy quadratic computational complexity. Extensive experiments demonstrate that ScopeViT achieves competitive complexity/accuracy trade-offs compared to existing networks across a wide range of visual tasks. On the ImageNet-1K dataset, ScopeViT achieves a top-1 accuracy of 81.1%, using only 7.4M parameters and 2.0G FLOPs. Our approach outperforms Swin (ViT-based) by 1.9% accuracy while saving 42% of the parameters, outperforms MobileViTv2 (Hybrid-based) with a 0.7% accuracy gain while using 50% of the computations, and also beats ConvNeXt V2 (ConvNet-based) by 0.8% with fewer parameters. Our code is available on GitHub.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
更新
PDF的下载单位、IP信息已删除 (2025-6-4)

科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
奋斗飞薇发布了新的文献求助10
1秒前
Eeeee发布了新的文献求助30
1秒前
红豆完成签到,获得积分20
1秒前
1秒前
1秒前
livian关注了科研通微信公众号
1秒前
2秒前
2秒前
2秒前
2秒前
2秒前
2秒前
开朗的风华完成签到,获得积分20
2秒前
2秒前
斯文败类应助科研通管家采纳,获得10
2秒前
2秒前
2秒前
2秒前
tuanheqi应助科研通管家采纳,获得10
2秒前
2秒前
2秒前
2秒前
tuanheqi应助科研通管家采纳,获得70
2秒前
2秒前
2秒前
2秒前
天天快乐应助科研通管家采纳,获得10
2秒前
2秒前
3秒前
3秒前
老大蒂亚戈应助月儿采纳,获得10
3秒前
3秒前
3秒前
4秒前
酷波er应助YYQ采纳,获得10
4秒前
龙傲天发布了新的文献求助10
4秒前
Saluzi发布了新的文献求助10
4秒前
5秒前
hkh发布了新的文献求助10
5秒前
高分求助中
【提示信息,请勿应助】关于scihub 10000
The Mother of All Tableaux: Order, Equivalence, and Geometry in the Large-scale Structure of Optimality Theory 3000
Social Research Methods (4th Edition) by Maggie Walter (2019) 2390
A new approach to the extrapolation of accelerated life test data 1000
北师大毕业论文 基于可调谐半导体激光吸收光谱技术泄漏气体检测系统的研究 390
Phylogenetic study of the order Polydesmida (Myriapoda: Diplopoda) 370
Robot-supported joining of reinforcement textiles with one-sided sewing heads 360
热门求助领域 (近24小时)
化学 材料科学 医学 生物 工程类 有机化学 生物化学 物理 内科学 纳米技术 计算机科学 化学工程 复合材料 遗传学 基因 物理化学 催化作用 冶金 细胞生物学 免疫学
热门帖子
关注 科研通微信公众号,转发送积分 4009834
求助须知:如何正确求助?哪些是违规求助? 3549753
关于积分的说明 11303647
捐赠科研通 3284309
什么是DOI,文献DOI怎么找? 1810591
邀请新用户注册赠送积分活动 886367
科研通“疑难数据库(出版商)”最低求助积分说明 811406