Skip to main content
Communities & Collections
All of DSpace
Statistics
English
العربية
বাংলা
Català
Čeština
Deutsch
Ελληνικά
Español
Suomi
Français
Gàidhlig
हिंदी
Magyar
Italiano
Қазақ
Latviešu
Nederlands
Polski
Português
Português do Brasil
Srpski (lat)
Српски
Svenska
Türkçe
Yкраї́нська
Tiếng Việt
Log In
Log in
New user? Click here to register.
Have you forgotten your password?
Home
理學院
資訊工程學系
學位論文
學位論文
Permanent URI for this collection
http://rportal.lib.ntnu.edu.tw/handle/20.500.12235/73912
Browse
Search
By Issue Date
By Author
By Title
By Subject
By Subject Category
Search
By Issue Date
By Author
By Title
By Subject
By Subject Category
1 results
Back to results
Filters
Author
search.filters.author.Lin, Yung-Hung
1
search.filters.author.林詠閎
Subject
1
search.filters.subject.Instance segmentation
1
search.filters.subject.Local augmentation
1
search.filters.subject.Object detection
1
search.filters.subject.Self-supervised learning
1
search.filters.subject.實例分割
Show more
Search subject
Submit
Browse subject tree
Date
Start
End
Submit
2025
1
Has files
1
No
Reset filters
Settings
Sort By
Accessioned Date Descending
Most Relevant
Title Ascending
Date Issued Descending
Results per page
1
5
10
20
40
60
80
100
Search
Author: search.filters.author.Lin, Yung-Hung
×
Search Tools
Search Results
Now showing
1 - 1 of 1
No Thumbnail Available
Item
具泛化能力的槽位表示之自監督學習
(
2025
)
林詠閎
;
Lin, Yung-Hung
Show more
自監督的發展在近年來受到了巨大的關注,其中對比式學習透過拉近同一圖像的另一視圖並推遠來自其他圖像的視圖,從而從未具標註的資料中學習表徵。近年以場景為主的影像資料集也開始被使用於預訓練,並且著重局部的學習可以在場景資料集表現得更好,這些方法大多依賴密集的匹配機制或是透過Selective Search找出可能物件,最近的方法透過將像素進行分群學習(稱作槽位),將相同語義的像素分配到同一槽位內,並讓習得的槽位可以隨著資料進行調整。我們發現全局的增強方法無法針對槽位調整,因此,我們提出了一種局部的特徵增強方法,透過對每個槽位進行特徵級別的增強,使槽位可以學習到資料的更多變化與型態,以提升泛化能力。我們在物件偵測、語義分割、多標籤分類等下游任務上評估我們所開發的自監督方法的性能,我們引入的方法不會增加訓練參數,並且在各個下游任務的表現上都有所提升。
Show more