Please enable JavaScript.
Coggle requires JavaScript to display documents.
Deep Visual Domain Adaptation: A Survey - Coggle Diagram
Deep Visual Domain Adaptation: A Survey
Abstract
Deep domain adaptation
相比較的方法
學習共享特徵子空間( learn shared feature subspaces)
使用淺層表示重用重要源實例(reuse important source instances with shallow representations)
目的
可以解決缺少大量標記數據的問題
方法
通過將域自適應嵌入到深度學習管道中來利用深層網絡來學習更多可傳遞表示(leverage deep networks to learn more transferable representations by embedding domain adaptation in the pipeline of deep learning)
論文貢獻
我們根據定義兩個域如何分開的數據屬性，提出了不同的深域適應方案的分類法(we present a taxonomy of different deep domain adaptation scenarios according to the properties of data that define how two domains are diverged)
我們基於訓練損失將深度域適應方法歸納為幾個類別，並簡要分析和比較這些類別下的最新方法(we summarize deep domain adaptation approaches into several categories based on training loss, and analyze and compare briefly the state-of-the-art methods under these categories)
我們概述了超出圖像分類的計算機視覺應用程序，例如人臉識別，語義分割和對象檢測(we overview the computer vision applications that go beyond image classification, such as face recognition, semantic segmentation and object detection)
突出了當前方法的一些潛在缺陷和未來的一些方向(some potential deficiencies of current methods and several future directions are highlighted)
Introduction
熟練地將輔助數據用於當前任務而缺乏數據將對現實世界的應用程序有所幫助(skillfully using the auxiliary data for the current task with scarce data will be helpful for real-world applications)
儘管公開提供的帶有標籤的大型視頻數據庫僅包含少量樣本(although large-scale labeled video databases that are publicly available only contain a small number of samples)
但從統計學上講，YouTube人臉數據集（YTF）包含3.4K視頻。 標記的靜止圖像數量足夠(statistically, the YouTube face dataset (YTF) consists of 3.4K videos. The number of labeled still images is more than sufficient)
由於許多因素（例如，照明，姿勢和圖像質量），這可能會降低性能(due to many factors (e.g., illumination, pose, and image quality) can degrade the performance)
模仿人類視覺系統的領域適應（DA）是轉移學習（TL）的特殊情況，它利用一個或多個相關源域中的標記數據在目標域中執行新任務(Mimicking the human vision system, domain adaptation (DA) is a particular case of transfer learning (TL) that utilizes labeled data in one or more relevant source domains to execute new tasks in a target domain)
DA
淺層DA( shallow DA)的常見算法
基於特徵的DA (feature-based DA )[37], [82], [30], [81]
通常學習一個公共共享空間，其中兩個數據集的分佈是匹配的(a common shared space is generally learned in which the distributions of the two datasets are matched)
基於實例的DA(instance-based DA) [6], [18]
通過對源樣本進行加權來減少差異，然後對加權後的源樣本進行訓練(reduces the discrepancy by reweighting the source samples, and it trains on the weighted source samples)
深層DA(deep DA)
[22]表明，域轉移仍然會影響其性能。 深層特徵最終將從一般性轉變為特定性，並且表示的可傳遞性在較高層中急劇下降([22] showed that a domain shift still affects their performance. The deep features would eventually transition from general to specific, and the transferability of the representation sharply decreases in higher layers)