# 02/06/25 Meeting Notes #20 # This Week's Progress 這週進度 ## Datasets - Made another category for tops as our current categories does not cover certain types of tops (singlets). Tops - 337 `為上衣創建了另一個類別,因為我們目前的類別不涵蓋某些類型的上衣(背心)` ## Classification Model ### Garment Classification We tried to remove the background for every image in the dataset and the result after removing the background is higher. `我們嘗試去除資料集中每張影像的背景,去除背景後的準確率更高。` | ![2fff27df-def9-4410-962e-cbba35a6b29f](https://hackmd.io/_uploads/H1ZnW0xtyg.jpg) | ![2fff27df-def9-4410-962e-cbba35a6b29f](https://hackmd.io/_uploads/H1lnWRxKke.jpg) | |:----------------------------------------------------------------------------------:| ---------------------------------------------------------------------------------- | | ![FINAL_load(89.44%)](https://hackmd.io/_uploads/ByjZGCgKyx.jpg) | ![FINAL_load(90.49%)](https://hackmd.io/_uploads/SyXzfAetye.jpg) | ### Style Classification We consider to use VTON(Virtual Try-on) model. It takes 2 inputs, garment image and a person. the output is the person wearing the garment clothes. `我們考慮使用VTON(Virtual Try-on)模型。它需要 2 個輸入:服裝圖像和人物。輸出是穿衣服的人` we have found the research paper (July 2024) and it has an open source code that can be used `我們找到了研究論文(2024 年 7 月),它有一個可以使用的開源程式碼`: research paper: https://arxiv.org/pdf/2403.05139 huggingface website (demo): https://huggingface.co/spaces/yisol/IDM-VTON github website: https://github.com/yisol/IDM-VTON we have tried running the code locally and this is the result`我們嘗試在本地運行程式碼,結果如下`: | Input | Output | |:------------------------------------------------:| :------: | |Garment image:![41](https://hackmd.io/_uploads/SkTm0MzYJx.png) |Masked image:![masked_output](https://hackmd.io/_uploads/HJl53RfzYyx.jpg)| |person image:![000205_0](https://hackmd.io/_uploads/H1eIJQGY1l.jpg)|result:![output](https://hackmd.io/_uploads/SylS17fK1e.jpg)| original idea: ![EMPS - dataflow(1)](https://hackmd.io/_uploads/H1DixQztkx.png) after implementing VTON: ![EMPS - dataflow (2)](https://hackmd.io/_uploads/SkDiemMFye.png) ## Frontend - Started discussing and designing the interface look. Currently incomplete. `開始討論和設計介面外觀。目前不完整。` [HERE LUCID CHART]( https://lucid.app/lucidchart/ab2d3896-f60c-413a-8551-af2983a1fed4/edit?viewport_loc=-2578%2C-180%2C4785%2C1989%2C0_0&invitationId=inv_2ca56601-2061-4dfc-a97d-7fd8af3898e4 ) # To Do 需做 - Upload the additional tops to github and run the model again. Need to determine the name of this category, please give me suggestion. And we add as class 14? or rename all the categories to have all tops in sequence? - Vton can be used if it helps our project more. - Continue designing the frontend. Determine a meeting time next week so we can discuss our preferences. - Proposal modification of the small details that were discussed with the professor. Submit to TAs before the 12th. # Next Weeks's Meeting 下週會議 --- Previous: Next: Full Content List [here](https://hackmd.io/@emps-113up/full-list)