Cynthia
    • Create new note
    • Create a note from template
      • Sharing URL Link copied
      • /edit
      • View mode
        • Edit mode
        • View mode
        • Book mode
        • Slide mode
        Edit mode View mode Book mode Slide mode
      • Customize slides
      • Note Permission
      • Read
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Write
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
    • Invite by email
      Invitee

      This note has no invitees

    • Publish Note

      Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

      Your note will be visible on your profile and discoverable by anyone.
      Your note is now live.
      This note is visible on your profile and discoverable online.
      Everyone on the web can find and read all notes of this public team.

      Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Explore these features while you wait
      Complete general settings
      Bookmark and like published notes
      Write a few more notes
      Complete general settings
      Write a few more notes
      See published notes
      Unpublish note
      Please check the box to agree to the Community Guidelines.
      View profile
    • Commenting
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
      • Everyone
    • Suggest edit
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
    • Emoji Reply
    • Enable
    • Versions and GitHub Sync
    • Note settings
    • Note Insights New
    • Make a copy
    • Transfer ownership
    • Delete this note
    • Save as template
    • Insert from template
    • Import from
      • Dropbox
      • Google Drive
      • Gist
      • Clipboard
    • Export to
      • Dropbox
      • Google Drive
      • Gist
    • Download
      • Markdown
      • HTML
      • Raw HTML
Menu Note settings Note Insights Versions and GitHub Sync Sharing URL Create Help
Create Create new note Create a note from template
Menu
Options
Make a copy Transfer ownership Delete this note
Import from
Dropbox Google Drive Gist Clipboard
Export to
Dropbox Google Drive Gist
Download
Markdown HTML Raw HTML
Back
Sharing URL Link copied
/edit
View mode
  • Edit mode
  • View mode
  • Book mode
  • Slide mode
Edit mode View mode Book mode Slide mode
Customize slides
Note Permission
Read
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Write
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
  • Invite by email
    Invitee

    This note has no invitees

  • Publish Note

    Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

    Your note will be visible on your profile and discoverable by anyone.
    Your note is now live.
    This note is visible on your profile and discoverable online.
    Everyone on the web can find and read all notes of this public team.

    Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Explore these features while you wait
    Complete general settings
    Bookmark and like published notes
    Write a few more notes
    Complete general settings
    Write a few more notes
    See published notes
    Unpublish note
    Please check the box to agree to the Community Guidelines.
    View profile
    Engagement control
    Commenting
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    • Everyone
    Suggest edit
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    Emoji Reply
    Enable
    Import from Dropbox Google Drive Gist Clipboard
       Owned this note    Owned this note      
    Published Linked with GitHub
    • Any changes
      Be notified of any changes
    • Mention me
      Be notified of mention me
    • Unsubscribe
    --- title: "【論文筆記】Convolutional Neural Networks for Fashion Classification and Object Detection" date: 2020-08-21 is_modified: false disqus: cynthiahackmd categories: - "智慧計算 › 人工智慧" tags: - "AI/ML" - "Fashion Classification" - "電腦視覺 CV" - "讀書筆記" - "文獻" --- {%hackmd @CynthiaChuang/Github-Page-Theme %} <br> > [Convolutional Neural Networks for Fashion Classification and Object Detection](http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf) > Lao B , Jagadeesh K(2015) > Final Paper, CS231N, Stanford <!--more--> ## 閱讀前自我提問 1. 期望能了解服裝分類領域中的 Know-how,包含但不限於:常用名詞、定義與方法、挑戰與現存的 benchmark。 ## 0. Abstract 1. **本文重點放在四個任務**: 1. 服裝類型(Type)分類 2. 服裝屬性(Attribute)分類 3. 相似服裝檢索 4. 服裝物體(Object)檢測 2. **結果**: - 在服裝風格(Style)分類上準確率有 50.0 % - 在服裝屬性(Attribute)分類上準確率有 74.5% 我這邊對於**相似服裝檢索**不感興趣,因此僅專注在服裝類型(Type)與屬性(Attribute)的分類上。是說,說到 Fashion Classification,可以瞄一下 ==Fashion MNIST== 資料集。 <p class="illustration"> <img src="https://i.imgur.com/oRSeVTN.png" alt=" Fashion MNIST 資料集"> Fashion MNIST 資料集(圖片來源: <a href="https://github.com/zalandoresearch/fashion-mnist">zalandoresearch/ fashion-mnist | github </a>) </p> ## 1. Introduction 段落一開始先介紹 Fashion classification 的應用: 1. 在**電子商務**方面,可以根據服裝照片推薦相似相品,或是推薦設計師作品。 2. 在**監視環境**中,可以利用行人服裝屬性輔助進行行人再識別(ReID) 3. 在**檢索應用**方面,可以使用文字檢索圖片,例如:穿紅衣的小女孩…等 <p class="illustration"> <img src="https://i.imgur.com/HJYJ70U.png" alt="洋裝下擺與單裙下擺"> 洋裝下擺與單裙下擺(左:洋裝、右:單裙)(圖片來源: 左: <a href="https://tw.baddiary.com/product/%E5%BE%8C%E6%8B%89%E9%8D%8A%E9%96%8B%E8%A1%A9%E8%A2%96%E5%A2%8A%E8%82%A9%E7%99%BE%E8%A4%B6%E6%B4%8B%E8%A3%9D%E9%99%84%E8%85%B0%E5%B8%B6/16754/?cafe_mkt=google_tw_dy&utm_source=Google&utm_medium=cpc&utm_campaign=Google_shopping&gclid=Cj0KCQjwg8n5BRCdARIsALxKb94BR8R5d_GGycXTqWv9UyWr_UdzFvFKTQN6VJ4T91wxmY8VS9dNkwYaAgxiEALw_wcB">baddiary</a>、右:<a href="https://www.google.com/url?sa=i&url=https%3A%2F%2Fworld.taobao.com%2Fproduct%2F%25E5%2596%25AE%25E8%25A3%2599%25E9%25A1%25AF%25E7%2598%25A6.htm&psig=AOvVaw2Nentv0o4g8jWz-gOhqw-y&ust=1597222436780000&source=images&cd=vfe&ved=0CA0QjhxqFwoTCKCcx7LjkusCFQAAAAAdAAAAABAI">淘寶海外</a>) </p> 並說明了在此領域會遇到的挑戰: 1. 各類衣服會具有==相似的特徵==,如:洋裝下擺與單裙下擺。 2. 服裝由於材料的的緣故,衣服容易變形,導致==特徵縮放==。 3. 視角與長寬比的不同,容易導致==衣物特徵變形==,而看起來不同。 ## 2. Problem Statement 在 [Abstract](#Abstract) 中出現的四種分類 Type、Attribute、Object 與 Style,這邊先進行定義。不過 Type 並沒有標註在圖上,該詞僅用於相似服裝檢索,根據上下文推測指的應該是==服裝的風格==,如:淑女、學院、中性、蘿莉塔、街頭、簡約…等。 <p class="illustration"> <img src="https://i.imgur.com/Qg6EdSw.png" alt="分類任務的摘要"> 分類任務的摘要(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> 這篇的四個子任務主要使用 CNN 來實做,不過他這邊提到 **Fashion classification has more generally consisted of nonCNN approaches**?這讓我感到有點驚訝,剛剛在看 Pedestrian Attribute Recognition(PAR),兩者都是在辨識身上的屬性,但在 PAR 中有大半的方法引入 CNN。 :::info :information_source: **關於 CNN PAR 的時間軸** Oops! 我好像搞混時間軸了。這篇是 2015 的文章,另外一篇被認為是是第一篇的 CNN PAR 論文 - Multi-attribute learning for pedestrian attribute recognition in surveillance scenarios,發表的時間也是 2015,是同一時期的事情。 ::: ### 2.1 Clothing Type Classification 服裝類型(Type)分類,是屬於 ==multiclass classification==,訓練時使用了 [Apparel Classification with Style (ACS)](https://data.vision.ee.ethz.ch/cvl/lbossard/accv12/) 資料集進行訓練,該資料集包含了 89,484 張圖片,主要是集中==上半身==衣物的分類,共有 15 個類別。 <p class="illustration"> <img src="https://i.imgur.com/sfmFXHQ.png" alt="ACS Dataset"> ACS Dataset(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> ### 2.2 Clothing Attribute Classification :::warning :warning: **the multi-label CNN architecture we are using** 同一段落中有提到這句話,作者認為所使用的是 multilabel?但在一個章節看網路架構圖,比較偏向我所認知的 **Multitask Learning** 的架構。 ::: 服裝屬性分類就是分類服裝==顏色==、==圖案==、==長度==…等屬性的問題。這些屬性有些可以用二進制量來表達,如:有無領帶;但有些無法,如:上裝顏色。感覺資料格式有點類似之前在看的 [Multitask classification](/@CynthiaChuang/Difference-between-Multiclass-Multilabel-and-Multitask-Problem#Multitask-classification)。 這部份網路訓練時採用的資料集是 [Clothing Attribute (CA) Dataset](http://chenlab.ece.cornell.edu/people/Andy/publications/ECCV2012_ClothingAttributes.pdf),該資料集包含了 1856 張==上身服裝==照片,,共有 26 個類別。 <p class="illustration"> <img src="https://i.imgur.com/rAQeZh7.png" alt="CA Dataset"> CA Dataset(圖片來源: <a href="http://chenlab.ece.cornell.edu/people/Andy/publications/ECCV2012_ClothingAttributes.pdf">論文</a>) </p> ### 2.3. Clothing Retrieval Pass ### 2.4. Clothing Object Detection 服裝物體檢測,顧名思義就是找出圖片中衣服存在的區域,就像下圖這樣: <p class="illustration"> <img src="https://i.imgur.com/ENlRAdX.png" alt="Clothing Object Detection"> Clothing Object Detection(圖片來源: <a href="https://arxiv.org/pdf/1807.01394.pdf">ModaNet</a>) </p> 這邊資料集採用 [Colorful-Fashion(CF) dataset](https://ieeexplore.ieee.org/document/6630093),但該資料集是 superpixel-labeling,所以在使用前需要先將標籤轉換為 ground-truth bounding box。這邊採用 Selective Search,且 intersection over union (IOU) 設為 0.5 來做標籤的處理。 <p class="illustration"> <img src="https://i.imgur.com/eOPeWwL.png" alt="CF Dataset"> CF Dataset(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> 這邊題到了幾個我不太熟的幾個名詞,稍微記錄一下: 1. **Superpixel** 中文翻作超像素,它是將一系列位置相鄰且顏色、紋理、亮度相似的像素組成的一個區域。這些小區域大多保留可供圖片的做進一步分割的資訊,因此可視為對圖片做基本資訊的抽象。最終所得的圖片會從一張像素級(pixel-level)的圖,變成區域級(district-level)的圖。 <p class="illustration"> <img src="https://i.imgur.com/lThCHJY.jpg" alt="superpixel"> superpixel(圖片來源: <a href="https://stackoverflow.com/questions/50264064/gabor-filter-on-on-each-superpixel">stackoverflow</a>) </p> - **Selective Search** 是個 pixel-based 的圖像分割演算法,詳細執行步驟可以看看這篇[網誌](https://www.jianshu.com/p/99e121c3beb8)。 - **IoU(Intersection over Union)** 在目標檢測中,IoU 指的是一種衡量指標,是用來計算模型產生的 bounding box 與原先標記的 bounding box 的重疊率。簡單來說就是算框的準不準,一般來說分數大於 0.5 就可以視為不錯的結果。 ## 3. Technical Approach and Models 作者總共訓練了四個網路分進行四個任務 ### 3.1. Clothing Type Classification 這邊他直接採用 AlexNet 進行遷移學習,在這邊的 AlexNet 在論文中是使用 Caff 實做的版本故又稱做 CaffeNet。並使用 ACS 資料集共 89,484 張、15個類別進行 Fine-Tune。 ### 3.2. Clothing Attribute Classification 屬性分類的部份,它使用 AlexNet 作為基礎,後接 26 個 Softmax 層,進行分類。標準的 Multi-task learning 的網路架構。 <p class="illustration"> <img src="https://i.imgur.com/ZXCao4H.png" alt="Clothing Attribute Classification"> Clothing Attribute Classification(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> ### 3.3. Clothing Retrieval Pass ### 3.4. Clothing Object Detection 物件偵測的部份採用 R-CNN 做遷移學習。並用改標註 ground-truth bounding box 的 CF 資料集進行 Fine-Tune。是說有點奇怪,是作者敘述順序放錯嗎?Object Detection 怎會在最後一步? ## 4. Results ### 4.1. Clothing Type Classification 在 [ACS 論文](https://data.vision.ee.ethz.ch/cvl/lbossard/accv12/accv12_apparel-classification-with-style.pdf)中,依照特徵提取分法的不同其準確率分別落在 35.03%、 38.29% 與 41.36%。而在本文中 Fine-Tune Full-Connected 與所有的層後,所的的準確率分別 46.0% 和 50.2%,高於論文原始數據。 <p class="illustration"> <img src="https://i.imgur.com/iyhFKam.png" alt="與 ACS 比較服裝分類結果"> 與 ACS 比較服裝分類結果(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> ### 4.2. Clothing Attribute Classification 分類結果看來對於顏色的分類是較為準確,但是於服裝細節的分類略糟。 <p class="illustration"> <img src="https://i.imgur.com/YCffx0m.png" alt="分類結果"> 分類結果(圖片來源: <a href="http://cs231n.stanford.edu/reports/2015/pdfs/BLAO_KJAG_CS231N_FinalPaperFashionClassification.pdf">論文</a>) </p> 是說作者舉了 placket 跟 solid 的例子,我還認真去查了下這兩個到底是什樣子。 <p class="illustration"> <img src="https://i.imgur.com/JC1ESan.png?1" alt=" placket 跟 solid"> 左: placket 右:solid(圖片來源: 左: <a href="https://www.beautifulhalo.com/mens-simple-long-sleeves-half-button-placket-loose-fit-solid-color-linen-shirt-p-407928.html">Beautifulhalo</a>、右:<a href="https://www.google.com/url?sa=i&url=https%3A%2F%2Fshopee.tw%2F%25E5%259B%259E%25E8%25B3%25BC%25E7%258E%2587%25E8%25B6%2585%25E9%25AB%2598%25EF%25BC%258C%25E8%25B6%2585%25E4%25BA%25BA%25E6%25B0%25A3%25E6%258E%25A8%25E8%2596%25A6-%25E7%2599%25BD%25E8%25A5%25AF%25E8%25A1%25AB-%25E7%2594%25B7-(%25E5%258F%25B0%25E7%2581%25A3%25E7%258F%25BE%25E8%25B2%25A8-%25E6%259C%2589%25E5%258F%25A3%25E8%25A2%258B)%25E4%25B8%258A%25E7%258F%25AD%25E6%2597%258F%25E7%2599%25BD%25E8%25A5%25AF%25E8%25A1%25AB-%25E9%2595%25B7%25E8%25A2%2596-mcps20-i.6368458.662654313&psig=AOvVaw3v5kCwmoCUStkFuUqkBW-1&ust=1597301305337000&source=images&cd=vfe&ved=0CAIQjRxqFwoTCLjDl5qJlesCFQAAAAAdAAAAABAD">蝦皮購物</a>) </p> ### 4.3. Clothing Retrieval Pass ### 4.4. Clothing Object Detection 在使用 R-CNN 做遷移學習時,做了兩階段的 Fine-Tune。在第一階段 Accuracy 達 91.25%、第二階段 達 93.4% ## 5. Discussion 簡略記錄下幾點: 1. 當圖片具有許多重疊特徵時,手動標記圖像可能會涉及一些主觀性,導致分類失誤。 2. 屬性分類中,目前顏色方面表現很好,但涉及細微屬性時,就有代加強。如果能建立一個更大的資料集,或許將能幫助模型學習。 ## 參考資料 1. Linear_Luo (2016-09-19)。[超像素(Superpixel)理解](https://blog.csdn.net/Linear_Luo/article/details/52588515) 。檢自 Linear_Luo的专栏 | CSDN博客 (2020-08-12)。 2. studyeboy (2019-06-28)。[超像素—学习笔记](https://blog.csdn.net/studyeboy/article/details/93981017) 。檢自 studyeboy的专栏 | CSDN博客 (2020-08-12)。 3. hanranV (2016-08-05)。[检测评价函数 intersection-over-union ( IOU )](https://blog.csdn.net/Eddy_zheng/article/details/52126641) 。檢自 hanranV的专栏 | CSDN博客 (2020-08-12)。 ## 更新紀錄 :::spoiler 最後更新日期:2020-08-21 - 2020-08-21 發布 - 2020-08-12 完稿 - 2020-07-31 起稿 ::: {%hackmd @CynthiaChuang/Github-Page-Footer %}

    Import from clipboard

    Paste your markdown or webpage here...

    Advanced permission required

    Your current role can only read. Ask the system administrator to acquire write and comment permission.

    This team is disabled

    Sorry, this team is disabled. You can't edit this note.

    This note is locked

    Sorry, only owner can edit this note.

    Reach the limit

    Sorry, you've reached the max length this note can be.
    Please reduce the content or divide it to more notes, thank you!

    Import from Gist

    Import from Snippet

    or

    Export to Snippet

    Are you sure?

    Do you really want to delete this note?
    All users will lose their connection.

    Create a note from template

    Create a note from template

    Oops...
    This template has been removed or transferred.
    Upgrade
    All
    • All
    • Team
    No template.

    Create a template

    Upgrade

    Delete template

    Do you really want to delete this template?
    Turn this template into a regular note and keep its content, versions, and comments.

    This page need refresh

    You have an incompatible client version.
    Refresh to update.
    New version available!
    See releases notes here
    Refresh to enjoy new features.
    Your user state has changed.
    Refresh to load new user state.

    Sign in

    Forgot password
    or
    Sign in via Google Sign in via Facebook Sign in via X(Twitter) Sign in via GitHub Sign in via Dropbox Sign in with Wallet
    Wallet ( )
    Connect another wallet

    New to HackMD? Sign up

    By signing in, you agree to our terms of service.

    Help

    • English
    • 中文
    • Français
    • Deutsch
    • 日本語
    • Español
    • Català
    • Ελληνικά
    • Português
    • italiano
    • Türkçe
    • Русский
    • Nederlands
    • hrvatski jezik
    • język polski
    • Українська
    • हिन्दी
    • svenska
    • Esperanto
    • dansk

    Documents

    Help & Tutorial

    How to use Book mode

    Slide Example

    API Docs

    Edit in VSCode

    Install browser extension

    Contacts

    Feedback

    Discord

    Send us email

    Resources

    Releases

    Pricing

    Blog

    Policy

    Terms

    Privacy

    Cheatsheet

    Syntax Example Reference
    # Header Header 基本排版
    - Unordered List
    • Unordered List
    1. Ordered List
    1. Ordered List
    - [ ] Todo List
    • Todo List
    > Blockquote
    Blockquote
    **Bold font** Bold font
    *Italics font* Italics font
    ~~Strikethrough~~ Strikethrough
    19^th^ 19th
    H~2~O H2O
    ++Inserted text++ Inserted text
    ==Marked text== Marked text
    [link text](https:// "title") Link
    ![image alt](https:// "title") Image
    `Code` Code 在筆記中貼入程式碼
    ```javascript
    var i = 0;
    ```
    var i = 0;
    :smile: :smile: Emoji list
    {%youtube youtube_id %} Externals
    $L^aT_eX$ LaTeX
    :::info
    This is a alert area.
    :::

    This is a alert area.

    Versions and GitHub Sync
    Get Full History Access

    • Edit version name
    • Delete

    revision author avatar     named on  

    More Less

    Note content is identical to the latest version.
    Compare
      Choose a version
      No search result
      Version not found
    Sign in to link this note to GitHub
    Learn more
    This note is not linked with GitHub
     

    Feedback

    Submission failed, please try again

    Thanks for your support.

    On a scale of 0-10, how likely is it that you would recommend HackMD to your friends, family or business associates?

    Please give us some advice and help us improve HackMD.

     

    Thanks for your feedback

    Remove version name

    Do you want to remove this version name and description?

    Transfer ownership

    Transfer to
      Warning: is a public team. If you transfer note to this team, everyone on the web can find and read this note.

        Link with GitHub

        Please authorize HackMD on GitHub
        • Please sign in to GitHub and install the HackMD app on your GitHub repo.
        • HackMD links with GitHub through a GitHub App. You can choose which repo to install our App.
        Learn more  Sign in to GitHub

        Push the note to GitHub Push to GitHub Pull a file from GitHub

          Authorize again
         

        Choose which file to push to

        Select repo
        Refresh Authorize more repos
        Select branch
        Select file
        Select branch
        Choose version(s) to push
        • Save a new version and push
        • Choose from existing versions
        Include title and tags
        Available push count

        Pull from GitHub

         
        File from GitHub
        File from HackMD

        GitHub Link Settings

        File linked

        Linked by
        File path
        Last synced branch
        Available push count

        Danger Zone

        Unlink
        You will no longer receive notification when GitHub file changes after unlink.

        Syncing

        Push failed

        Push successfully