README.md

metadata

language:
  - en
license: other
task_categories:
  - image-classification
  - multiple-choice
  - visual-question-answering
tags:
  - fashion
  - recommendation-system
  - polyvore
pretty_name: Polyvore Outfits (Refactored)
configs:
  - config_name: disjoint_compatibility
    data_files:
      - split: train
        path: disjoint_compatibility/train.json
      - split: validation
        path: disjoint_compatibility/valid.json
      - split: test
        path: disjoint_compatibility/test.json
  - config_name: disjoint_fill_in_the_blank
    data_files:
      - split: train
        path: disjoint_fill_in_the_blank/train.json
      - split: validation
        path: disjoint_fill_in_the_blank/valid.json
      - split: test
        path: disjoint_fill_in_the_blank/test.json
  - config_name: nondisjoint_compatibility
    data_files:
      - split: train
        path: nondisjoint_compatibility/train.json
      - split: validation
        path: nondisjoint_compatibility/valid.json
      - split: test
        path: nondisjoint_compatibility/test.json
  - config_name: nondisjoint_fill_in_the_blank
    data_files:
      - split: train
        path: nondisjoint_fill_in_the_blank/train.json
      - split: validation
        path: nondisjoint_fill_in_the_blank/valid.json
      - split: test
        path: nondisjoint_fill_in_the_blank/test.json

Polyvore Outfits (Refactored Version)

This repository provides a refactored version of the Polyvore Outfits dataset, originally introduced in the paper "Learning Type-Aware Embeddings for Fashion Compatibility" by Mariya I. Vasileva et al.

📌 Overview

The goal of this refactoring is to improve usability and developer experience. While the core data remains identical to the original, the file structure and JSON schemas have been standardized to make it easier to load and use in modern deep learning pipelines.

📊 Data Formats

1. Compatibility

Binary classification data to determine if a set of items matches.

• label: 1: Compatible
• label: 0: Incompatible

{
  "items": ["217320763_1", "217320763_2"],
  "label": 1
}

2. Fill-in-the-Blank (FITB)

Multiple-choice data to predict the most suitable item for a missing slot.

{
  "items": ["217320763_1", "217320763_2"],
  "blank_index": 2,
  "candidates": ["217320763_3", "xxxx_x"],
  "label": 0
}

📜 Citation & Original Work

This is a refactored version of the work by Vasileva et al. If you use this dataset, please cite the original paper:

@misc{vasileva2018learningtypeawareembeddingsfashion,
      title={Learning Type-Aware Embeddings for Fashion Compatibility},
      author={Mariya I. Vasileva and Bryan A. Plummer and Krishna Dusad and Shreya Rajpal and Ranjitha Kumar and David Forsyth},
      year={2018},
      eprint={1803.09196},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={[https://arxiv.org/abs/1803.09196](https://arxiv.org/abs/1803.09196)},
}

⚖️ License

This refactored version follows the same licensing terms as the original Polyvore Outfits dataset. Please refer to the original source for specific usage restrictions.