README.md

Disclaimer: This repository serves for inspirational purposes only.

Style2Vec

Implementation of fashion products vector representation based on Style2Vec: Representation Learning for Fashion Items from Style Sets.

Based on the intuition of distributional semantics used in word embeddings, Style2Vec learns the representation of a fashion item using other items in matching outfits as context. Two different convolutional neural networks are trained to maximize the probability of item co-occurrences.

Docs

• Report

Data Sources

• Polyvore Dataset was used for model training.
• DeepFashion: Attribute Prediction was used for model evaluation

Requirements

• Tensorflow > 1.14
• Matplotlib
• scikit-learn
• Pillow
• scipy
• pandas

Results

Sequence of nearest neighbors from Style2Vec embedding on each line:

Project Organization

├── README.md          <- The top-level README for developers using this project.
|
├── data
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- Project documentation and reports
│   └── report.md      <- Project report
│
├── models             <- Trained and serialized models, model predictions, or model summaries
|
└── style2vec          <- Source code for use in this project.
    ├── data           <- Scripts to process data
    │   │
    │   ├── deepfashion_prep.py <- DF dataset preprocessing
    |   ├── sample_generator.py <- Samples generator for model training
    │   └── preprocessing.py    <- Image preprocessing
    │
    ├── features       <- Scripts to turn raw data into features for modeling
    |   ├── df_embedding.py       <- Deepfashion dataset embedding
    │   └── polyvore_embedding.py <- Polyvore dataset embedding
    │
    ├── models         <- Scripts to train models and then use trained models to make predictions
    │   └── style2vec.py <- Style2Vec model training
    │
    └── visualization  <- Scripts to create exploratory and results oriented visualizations
        │
        ├── df_attr_comparison.py <- Visualization of nearest negihbors with their attributes
        ├── df_exploration.py     <- Deep Fashion dataset statistical exploration
        ├── df_neighbors.py       <- Model validation
        ├── n_neighbors.py        <- sk-learn n-neighbors wrapper
        └── polyvore_neighbors.py <- Nearest neighbors visualization of Polyvore dataset embedding