Contents

1. Introduction

2. Start the app

3. Repository structure

1. Intro

Dread. Dilemma. Distraction. Picking out a daily outfit from your wardrobe often involves these emotions. Given the abundance of clothes in our wardrobes, picking an outfit should be simpler and quicker.

Could this be a task our mobiles can do as well? We set out exploring this. Our idea is to apply computer vision based deep learning techniques to recommend outfits. We develop a local app that can be accessed on our systems which identifies outfits as a combination of 3 subcategories - Topwear, Bottomwear and Footwear. Furthermore, we give you the outfits based on the season of the year. You may choose to look ahead for a specific time in the future to plan your outfit in advance.

The dataset used for our project is Fashion Product Images (small) which you can find here. The dataset contains training images in the images folder, a styles.csv file, and test images in the test folder.

2. Start the app

To start the app:

a) Check whether saved models are present inside the models/models/ directory. There should be in total 4 models, one inside each subdirectory for category, topwear, bottomwear and footwear. You can find the trained models here.

b) Check whether all dependencies are installed using

   pip install -r requirements.txt

c) :aunch the system UI using

   python3 ui_module.py   

3. Repository structure

.
├── data
│   ├── images # A directory containing the dataset images
│   └── styles.csv # A csv file containing the annotations for the images
├── models # A directory containing trained models and images of the model architectures
│   ├── models # A google drive folder which contains our trained models
│   │   ├── model_category # A model that distinguishes tops, bottoms, and shoes
│   │   ├── model_bottomwear # A model that recognizes the type, color, gender, season, and usage of bottoms
│   │   ├── model_footwear # A model that recognizes the type, color, gender, season, and usage of shoes
│   │   └── model_topwear # A model that recognizes the type, color, gender, season, and usage of tops
│   ├── model_bottomwear.png # Architecture of the model_bottomwear
│   ├── model_category.png # Architecture of the model_category
│   ├── model_footwear.png # Architecture of the model_footwear
│   └── model_topwear.png # Architecture of the model_topwear
├── ui_images # A directory containing images used for the ui_module
├── exploratory_data_analysis.ipynb # A notebook containing the analysis of our dataset and corresponding inferences
├── model_category.ipynb # A notebook containing the training of the model_category
├── model_category_hyperparam.ipynb # A notebook containing code for hyperparameter tuning for model_category
├── models_subcategory.ipynb # A notebook containing the training of the model_bottomwear, model_footwear, and model_topwear
├── models_subcategory_hyperparam.ipynb # A notebook containing code for the hyperparameter tuning for subcategory models
├── recognition_module.py # A module that contains functions and classes to generate the GUI#
├── ui_module.py # A module to run the application
├── utils.py # A module containing helping functions for model training
├── WardrobeAI_report.pdf # Project report
├── README.md # The Readme file
└── requirements.txt # The packages used