# 9/19/24 Meeting Notes #5

# This week's progress 這週進度
- Classification modeul generative modules. 分類和生成模組
- Defined main style categories and subcategories 定義主要風格類別和子類別

## Modules
- Two modules for classification and one module for generative. 兩個分類模組和一個生成模組
- **First classification module**
-- Classifies user's wardrobe into top, bottom, shoes, socks, accessories. 將使用者的衣櫃分類為上衣、下裝、鞋子、襪子、配件。
`Example: After user takes a picture of a shirt, first confirmation will be to ask user to confirm if it is a shirt. Second confirmation is asking the fit of the shirt (oversized tee, regular fit, crop top, etc.), in order to classify the item more accurately. 使用者拍攝一件襯衫的照片，系統會首先要求使用者確認這是否是一件襯衫。第二次確認是詢問襯衫的版型（寬鬆版T恤、常規版型、短版上衣等），以便更準確地分類該物品。`
- **Second classification module**
-- Classifies outfit into appropriate styles. 將穿搭分類為適當的風格。
-- First idea: to define a style by the attribute of the items, how it fits, and its material. 根據物品的屬性、版型及材質來定義風格。
`Example: user wants an outfit with "A" style. This will need "A" fit of top, bottom, shoes, etc. Then we can just show those attrivutions and exclude items that do not have "A" attributes. 使用者想要一套“A”風格的穿搭。這需要“A”風格的上衣版型、下裝、鞋子等。然後我們可以只顯示符合“A”屬性的物品，排除不符合的物品。`
-- Second idea: Using brute force to get different outfits.
-- Third idea: Update the new outfit combination and classify it to a style, every time there is a new item added. 每當添加新物品時，更新新的穿搭組合並將其分類為一種風格。

- **Generative module**
-- Generate a 2D/3D model with the outfit from the output of the second classification module. 使用第二個分類模組的輸出生成服裝的 2D/3D 模型。

## Style categories
- Casual 休閒
-- Beach 海灘風
-- Comfy 舒適風
-- Edgy 前衛風
- Formal 正式
-- Business 商務風
-- Evening 晚宴風
- Semi-Formal 半正式
-- Preppy 學院風
-- Business casual 商務休閒
-- Classic 經典風
- Sporty 運動風
- Vintage 復古
-- Boho 波希米亞風
-- Streetwear 街頭風 (→ Punk 龐克, → Hip Hop 嘻哈/ Y2K)

## Questions
- Is the definition of style depending on the colour, material, and fitment (loose/regular/fit/crop) enough? 僅依賴顏色、材質和版型（寬鬆/常規/合身/短版）來定義風格是否足夠？

- We would like to change the recommendation model into the first classification module. Since the recommendation module is a side feature for our project. Would that be okay for the technical contribution of this project? If we got enough time to build the recommendation model, maybe we could make it.我們想把推薦模組改成第一個分類模組。由於推薦模型是我們project的一個附帶功能,這樣對於這個project的技術貢獻可以嗎？如果我們有足夠的時間來建立推薦模型，也許我們可以再加進去。

- Another idea for the side feature that uses deep learning is AI chatbot. The idea of this chatbot is to be the advisor for the user. Lets say the user want to go to an event but they dont know what style should they choose, they can ask to the chatbot for some suggestions of what kind of style they can wear to go for a specific event.（輔助功能））另一個想法是把AI chatbot 加進去成為使用者的顧問。假設使用者想去參加一個活動，但他們不知道應該選擇什麼風格，他們可以向聊天機器人詢問一些關於他們可以穿什麼樣的風格去參加特定活動的建議。

# To do:
- Understand ETM module and image-to-image module.
- survey and compared papers. Best be papers with code.
- Find papers that can do image-to-image using a singular piece of clothing and a model. The find papers that can do image-to-image by using merge pieces of clothing and a model. (merged pieces = top + bottom + socks + shoes + accessories).
- Use either GAN or diffusion model to become our system.
- Figure out what each module do.

# Next week's meeting
- The deciding of ideas for second classification model move to next week (Haven't decided which idea is more suitable).
- Architure flow diagram of our system.
- ETM modules and image-to-image modules.

---
Previous: [9/12 Meeting Notes #4](https://hackmd.io/@emps-113up/meeting4)
Next: [9/26 Meeting Notes #6](https://hackmd.io/@emps-113up/meeting6)
Full Content List [here](https://hackmd.io/@emps-113up/full-list)