# 10/24/24 Meeting Notes #10
# This Week's Progress 這週進度
## Classification model#1
- No progress for this model 無進度
## Classification model #2
- The paper used both the DeepFashion and their own private datasets.
- `這篇論文使用了 DeepFashion 和他們自己的私有數據集`
-- We downloaded DeepFashion but couldn't use it because its labels didn't match what we needed.
`我們下載了 DeepFashion，但無法使用，因為其標籤與我們需要的不一致`
- We also downloaded FashionStyle14, a public dataset that classifies images of girls into different styles.
- `我們也下載了 FashionStyle14，這是一個將女性圖片分類為不同風格的公共數據集`
-- FashionStyle14 comes with three files: train.csv, test.csv, and val.csv, each containing image paths for training, testing, and validation datasets.
`FashionStyle14 附帶三個文件：train.csv、test.csv 和 val.csv，分別包含訓練、測試和驗證數據集的圖片路徑。`
-- Some image paths in these files don't exist, so we need to manually select or write a Python script to choose the correct images for training and testing.
`這些文件中的某些圖片路徑不存在，因此我們需要手動選擇或編寫 Python 腳本來挑選正確的訓練和測試圖片。`
-- We also need to oversample the dataset because some categories have far fewer images than others.
`我們還需要對數據集進行過度採樣，因為某些類別的圖片數量遠少於其他類別。`

237 openable files out of 1295 in conservative
55 openable files out of 1010 in dressy
548 openable files out of 563 in ethnic
110 openable files out of 1017 in fairy
133 openable files out of 198 in feminine
132 openable files out of 1086 in gal
805 openable files out of 2030 in girlish
268 openable files out of 1252 in kireime-casual
111 openable files out of 1119 in lolita
486 openable files out of 1489 in mode
205 openable files out of 1128 in natural
228 openable files out of 1269 in retro
280 openable files out of 424 in rock
57 openable files out of 1068 in street

- Currently downloading Fashion144k, Fashion144k(StyleNet), and Fashion550k. Will still need to check if its usable for classification model #2
`目前正在下載 Fashion144k、Fashion144k (StyleNet) 和 Fashion550k。仍需檢查這些數據集是否適用於分類模型 #2`

## Generative model
- We are moving the environment to a Docker container.
`正在將環境轉移到 Docker 容器中`
- Docker has been successfully installed on the computer.
`Docker 已經成功安裝在電腦上`
- Currently, we're facing issues with setting up the Dockerfile to configure the environment.
`目前，我們在設定 Dockerfile 來配置環境時遇到了一些問題`
- We're also looking for a way to install psbody, as it's not available on Docker Hub.
`我們也在尋找安裝 psbody 的方法，因為 Docker Hub 上沒有這個套件。`

## Dataset
- Divided more datasets from *Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback (CVPR 2021)*.
-- Individual dress piece 268 -> 465 (for classification model #1)
`單件連衣裙圖片從 268 增加到 465（用於分類模型 #1）`
-- Person wearing dress 1720 -> 2483 (Still need to divide into style categories for classification model #2)
`穿著連衣裙的人物圖片從 1720 增加到 2483（還需要進一步劃分成不同風格類別，用於分類模型 #2）`
- Dataset from *Dress Code: High-Resolution Multi-Category Virtual Try-On (ECCV 2022)*
-- Waiting for author to send dataset download link.
`正在等待作者發送數據集下載連結`
-- Dataset will be used to classification model #2
` 該數據集將用於分類模型 #2`
- Dataset from *Fashion 144k* is outdated (from 2015). Unable to use as fashion styles from back then no longer match the definition of styles today.
`Fashion 144k 數據集過時了（來自 2015 年）。由於當時的時尚風格已經不再符合當今的風格定義，無法使用。`

# To Do For Next Week 需做
- Use the other three datasets(Fashion144k, Fashion144k(StyleNet), and Fashion550k), merge and test if the code for classification model #2 is usable.
-- If the accuracy is still low, then the problem is with the code.
- Find papers that have models that divide the big categories then the small categories.
- Continue finding and dividing datasets.
- Finishing up the docker environment

---
Previous: [10/17/24 Meeting Notes #9](https://hackmd.io/@emps-113up/meeting9)
Next: [10/31/24 Meeting Notes #11](https://hackmd.io/@emps-113up/meeting11)
Full Content List [here](https://hackmd.io/@emps-113up/full-list)