# 11/21/24 Meeting Notes #13 # This Week's Progress 這週進度 ## Classification Model #1 - Previous testing the model with `fashionMNIST` dataset `之前用 FashionMNIST 資料集測試模型` (Training Accuracy: 90.39%, Testing Accuracy: 90.32%) - Trained the model with our own dataset. `使用我們自己的資料集訓練模型` (Training Accuracy: 92.88%, Testing Accuracy: 92.32%) ![image](https://hackmd.io/_uploads/Hyvmbv3Mkg.png) ![image](https://hackmd.io/_uploads/rkdtbw3G1x.png) - Trained Categories `分類項目`: ``` Dress Flats Heels Pants Shirt Shoes Shorts Skirt Sneakers Tshirt ``` - Test Dataset `測試資料集` ``` 756 files in dress 780 files in flats 780 files in heels 1029 files in pants 831 files in shirt 338 files in shoes 806 files in shorts 616 files in skirt 782 files in sneakers 877 files in tshirt total: 7595 ``` - Train Dataset `訓練資料集` ``` 3022 files in dress 3116 files in flats 3118 files in heels 4116 files in pants 3320 files in shirt 1349 files in shoes 3221 files in shorts 2463 files in skirt 3127 files in sneakers 3506 files in tshirt total: 30358 ``` - Currently missing dataset for polo shirt, sweaters and hoodies. `目前缺少有關 Polo 衫、毛衣和連帽衫的資料集。` - Datasets for hoodies -- Hoodies [HERE](https://www.kaggle.com/datasets/lmaoded11/hoodies) -- Clothing & Models (hoodies) [HERE](https://www.kaggle.com/datasets/dqmonn/zalando-store-crawl?resource=download) ## Classification Model #2 - Got the dataset from `Dress Code: High-Resolution Multi-Category Virtual Try-On (ECCV 2022)` `已獲取資料集自《Dress Code: High-Resolution Multi-Category Virtual Try-On (ECCV 2022)》` - `Dress Code` Paper [HERE](https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136680337.pdf) ![image](https://hackmd.io/_uploads/HyI-DLJg1x.png) -- Terms and conditions of the Licence [HERE](https://drive.google.com/file/d/1NA4zjng6ZHdNfFU7iL8Sf50o5A0Fy4E0/view?usp=sharing) - Dataset for training changed from `Fashion550k` to `Dress Code` `訓練使用的資料集已從 Fashion550k 更改為 Dress Code` - `Fashion550k` dataset is too outdated and background is messy. `資料集過於過時，且背景雜亂` - `Dress Code` dataset is professionally taken and much more suitable for our model. `資料集為專業拍攝，更適合我們的模型需求` - Modified style categories to fit our styles better. `已修改風格分類，使其更符合我們的風格定義` - Previous style categories`之前風格分類`: ``` 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) ``` - Updated style categories `最新風格分類` ``` Casual 休閒 - Beach 海灘風 - Comfy 舒適風 Formal 正式 Semi-Formal 半正式 – Preppy 學院風 – Business casual 商務休閒 – Classic 經典風 Sporty 運動風 Boho 波希米亞風 Streetwear 街頭風 ``` ## Generative Model - Docker is all set up for generative `已完成設定Docker 給生成模型` - Found a landmark module BCRNN [HERE](https://github.com/zuoxiang95/BCRNN) `找到BCRNN地標模組` - Waiting for generative dataset `等待生成模型的資料集` # To Do 需做 - Find and sort datasets for polo shirts, sweaters and hoodies for classification model #1 - Find datasets for generative model. --- Previous: [11/07/24 Meeting Notes #12](https://hackmd.io/@emps-113up/meeting12) Next: [11/21/24 Meeting #2 with Professor](https://hackmd.io/@emps-113up/meetingprof2) Full Content List [here](https://hackmd.io/@emps-113up/full-list)